Method and system for generating a desired concentration of functional food components in plants
The plant growing system optimizes environmental conditions to achieve desired functional food component concentrations in crops, addressing the lack of precision in existing systems and enhancing nutritional value without genetic modification.
Patent Information
- Application Number
- PCT/EP2026/063319
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-08
- Filing Date
- 2026-05-07
- Publication Date
- 2026-11-12
AI Technical Summary
Existing indoor growing systems lack the capability to precisely monitor and control the nutritional value of crops, particularly in terms of generating desired concentrations of functional food components, which are beneficial for human health, and do not allow for accurate influence of amino acid, mineral, or iron accumulation in plants.
A plant growing system with a sensor and actuator system controlled by a controller that follows a recipe to optimize environmental conditions for achieving desired functional food component concentrations, using a control algorithm to influence growth conditions such as temperature, humidity, CO2, light intensity, and nutrient availability.
Enables precise control of functional food component concentrations in plants, allowing for the development of crops with enhanced nutritional value without relying on genetic modification, thereby improving the healthiness of the end product.
Smart Images

Figure EP2026063319_12112026_PF_FP_ABST
Abstract
Description
[0001] P37107PC00
[0002] Method and system for generating a desired concentration of functional food components in plants
[0003] The present invention relates to indoor fresh food production, in particular to indoor growing of plants and / or crops.
[0004] The present invention relates to indoor fresh food production, in particular to indoor growing of plants. For indoor growing of crops, also referred to as Controlled Environmental Agriculture (CEA) or plant factories, indoor growing environments are used. In such indoor growing environments, growth conditions are measured and controlled such that plants can be grown efficiently year-round. An indoor growing environment is for example a greenhouse, a vertical farm, an indoor module, a container, or in general any growing space which is sheltered from the elements. The indoor growing environment comprises sensors for monitoring growth conditions during the growth cycle of the plant, and actuators for influencing, i.e., controlling, one or more of these growth conditions during the growth cycle of the plant. In indoor growing environments, different types of cultivation methods are employed. One such cultivation method is hydroponics, in which plants are grown using a water-based nutrient solution, rather than soil. Hydroponics allow for a precise monitoring and control of nutrients being delivered to the plant during the growth cycle of the plant.
[0005] In the remainder of this document, the following definitions are adhered to.
[0006] A growth condition is any condition of the indoor growing environment which affects the growing of the plant that is grown in the indoor growing environment. Examples of growth conditions are temperature, relative humidity, water supply, nutrient solution in the water, and lighting parameters such as light colour spectrum and light intensity.
[0007] A growth cycle of a plant comprises a plurality of growth stages to grow plants, which comprise at least of sowing seeds, or seedlings, and harvesting of the plant. The plant is planted in a medium, for example wherein the medium is a plug, such as e.g. commercially available GROWFOAM, Eazy Plug, or Nygaia plugs, or a block, such as e.g. a Rockwool block. Either a seed of the plant is directly planted in the medium in the indoor growing environment, or alternatively, the seed is allowed to germinate outside of the indoor growing environment, after which it is transferred to the medium as a seedling.
[0008] A growth stage is a point in time or a time period during the growth cycle in which the plant develops by, e.g., by growing leaves, performing photosynthesis, taking up nutrients, synthesizingbioactive ingredients, etc. During a growth stage, the actuators of the indoor growing environment control the growth conditions in order to influence the development of the plant.
[0009] A crop is a plant suitable for human consumption or for medical use. For example, the crop is a fruit-bearing plant or a leafy vegetable such as bok choy, lettuce, or basil.
[0010] CEA is a term used worldwide to describe the growth of plants / crops indoor in controlled growing conditions. Traditionally, the growth conditions in the indoor growing environment are either static throughout the growth cycle of the crop, or they are varied according to a “recipe”, the recipe specifying how the growth conditions during the growth cycle are varied. Such recipes may for example be derived from historical results of previous growth cycles of that plant. Typically, the different growth conditions are controlled during the growth cycle of the plant such that a particular objective is achieved. A control system is for example employed which monitors the growth conditions in the indoor growing environment by receiving sensor readings from sensors placed inside of the indoor growing environment, and by controlling actuators within the growing environment to influence the growth conditions such that they conform to the prescribed growth conditions in the recipe.
[0011] Several manufacturers of CEA control systems (i.e., for control of indoor growing environments) exist today. Their systems are discussed below.
[0012] Fluence - e.g., as described in patent application US20200134741 A1 - develops LED lighting solutions for indoor agriculture and a Platform-as-a-Service (PaaS) that monitors and controls various aspects of the growing environment, including illumination, irrigation, and fertilization. They utilize multispectral analysis to detect stress or disease in plants and adjust growing conditions accordingly. In particular, the following parameters of the plant growth are analysed:
[0013] Normalized Difference Vegetation Index (NDVI): determines the density of greenery in a given area by assessing light absorption and reflection patterns, often used to assess plant health.
[0014] CRI (Carotenoid Reflectance Index): measures the concentration of carotenoids in the plants, with higher levels often indicating plant stress.
[0015] Chlorophyll and Anthocyanin Indices: the concentration of these pigments provides insight into plant health and growth.
[0016] Source.ag - e.g., as described in patent application W02024028360A1 - develops Al-powered solutions for greenhouse optimization. Their system uses plant status data and long-term constraints to determine the optimal environmental actions for maximizing plant growth and yield. They employ machine learning models to predict plant responses and optimize greenhouse operations. The plant status data is “short-term data” collected by individual farmers / growers whichis combined with long-term constraints to then determine environmental actions that operate a greenhouse environment installation. To determine observable characteristics of the plant, a plant growth and development model is updated based on the plant status data, such as for example a number of leaves per stem, a size of the leaves, a density of stems, a number of fruits, a colour of fruits or a chemical composition of a leaf or fruit. The response of the system is based on an objective function which is extremised by estimating a value of the objective function using a machine learning model, the machine learning model being trained on simulations of the response of the one or more plants, or parts thereof, to the greenhouse environment and, optionally, plant actions performed on the one or more plants. The data for the decision model are based on averages.
[0017] Infinite Acres - e.g., as described in patent application US20210137028A1 - develops automated growing systems that utilize Al algorithms to monitor plant characteristics and environmental data, taking corrective actions to remedy plant health deficiencies. Their system continuously learns and adapts based on historical data, refining its control strategies overtime.
[0018] AvaGrow - e.g., as described in patent application US20200344965A1 - develops dynamic recipe management methods and systems for controlling fluid flow and lighting spectrums in controlled environments. Their system utilizes plant phenotyping data, obtained through image analysis, to adjust growing conditions based on plant type, growth stage, and other factors.
[0019] Hortau - e.g., as described in patent application W02020034039A1 - offers a system that monitors and regulates plant productivity using machine learning algorithms. Their system analyses data from sensors to predict crop variables and determine necessary interventions. They aim to optimize crop yield, resource use, and environmental impact.
[0020] Blue River Technology - e.g., as described in patent application US20200355829A1 -specializes in machine vision and robotics for agriculture. They have developed systems that use plant physiology sensors, including multispectral and hyperspectral cameras, to detect plant parameters and optimize growing conditions.
[0021] Ecoation - e.g., as described in patent application US20220104435A1 and US20170032258A1 - develops Al-powered systems for monitoring crop health and detecting early signs of stress or disease. Their systems use machine learning to determine ideal environmental conditions for expressing desired plant characteristics, for example shape, color, taste, size, and yield.
[0022] Blue Radix develops algorithm-based solutions for greenhouses. A crop controller optimizes the climate and irrigation within the greenhouse based on the growers’ crop strategy, weather data, and data received from sensors within the greenhouse.
[0023] In the paper “Eshkabilov, S.; Stenger, J.; Knutson, E.N.; Kuguktopcu, E.; Simsek, H.; Lee, C.W. Hyperspectral Image Data and Waveband Indexing Methods to Estimate NutrientConcentration on Lettuce (Lactuca sativa L.) Cultivars. Sensors 2022, 22, 8158”, the authors study the influence of Nitrogen (N), phosphorus (P), and potassium (K) compounds on the growth dynamics of four hydroponically grown lettuce cultivars. The NPK compounds were supplied as fertilizer to the plants. Hyperspectral images (HSI) were captured at plant harvest and used to estimate nutrient levels of leaf tissues. Such HSI data was shown to correlate well with laboratory-measured data. The cultivars were grown in hydroponic tubs using 8 different nutrient concentrations of N, P, and K. The cultivars were grown in a greenhouse at fixed 16-h day / 8-h night with a fixed light intensity and fixed temperatures (the growth “recipe” was thus fixed). Thus, the system described in this paper estimates concentrations of N, P, and K by using HSI captured at plant harvest. The estimation of these nutrient levels in the plant are used to determine optimal inputs of fertilizer in the water supply. The study does not address the development of functional food components during the growth cycle or at harvest. Instead, the paper estimates the uptake of N, P, and K compounds of the fertilizer into the plants. However, the paper does not disclose a precise measure of N, P, and K compounds, nor does it disclose any measures of the improvement in N, P, and K accumulation. Hence, said system also does not allow for a user to accurately influence the level of N, P, and K in the end product (i.e., the harvested plant).
[0024] In the paper “Khonh et al., Nanosensor for Fe(ll) and Fe(lll) Allowing Spatiotemporal Sensing in Planta, Nano Lett. 2025, 25, 6, 2316-2324”, a system which allows for real-time detection of iron in plants is described. Iron uptake (from fertilizer) is detected in the plant tissues or subcellular compartments. These changes in iron content may for example indicate how a plant handles stress and uses nutrients. The nanosensors are based on near-infrared (NIR) technology, using the Corona Phase Molecular Recognition (CoPhMoRe) technique that distinguish Fe(ll) and Fe(lll) species. The paper does not disclose any measures of the improvement in iron accumulation. Hence, the system described in the paper also does not allow for a user to accurately influence the level of iron in the end product (i.e., the harvested plant).
[0025] Present systems are thus generally directed to improving the health of plants during the growth cycle (e.g., avoiding stress or disease), crop yield, resource use, or environmental impact, or to detect nutrient uptake. Further, such systems are dependent on the local data gathered by the individual farmer / grower and sent to the system. The systems are directed to automating certain parts of the growing process within the indoor growing environment. Present systems thus allow for a more economical growing process when compared to a more traditional growing process which relies on the knowledge and manual interventions of the grower. Such present systems may thus increase yield, decrease waste, and reduce the manual labour involved in the growing process.There is a growing demand for transparency in nutritional value of food. Functional food components and bioactive components can contribute to a healthy diet and a healthy living. Ultraprecision CEA is the smart process of producing fresh food, such as plants, with a specific treatment during the growth cycle to develop desired concentrations of functional food components and bioactive components that is beneficial to the health of humans. In other words, with ultra-precision CEA, plants are produced that contain more nutrition (i.e., has a higher nutritional value). Functional food components for example belong to the following categories of bio-active components:
[0026] - Amino acids, Peptides, and Proteins;
[0027] Polyunsaturated fatty acids;
[0028] Vitamins and their analogues;
[0029] Minerals and trace elements;
[0030] Carotenoids;
[0031] - Antioxidants;
[0032] Dietary fiber;
[0033] Glucosinolates and Isothiocyanates;
[0034] Enzymes;
[0035] Bio-active components that can be extracted (such as in teas).
[0036] To influence these functional food components, several techniques are conceivable. With biofortification techniques, the concentration of mineral elements in a nutrient solution is adjusted to promote the accumulation of target minerals in the plant. Genetic biofortification strategies rely on genetic breeding techniques (such as GMO and Crispr gene editing) aimed at cultivating or optimising new plant varieties with enhanced mineral element accumulation capacities. In certain systems, such as in plant factories, carbon dioxide (CO2), nutrient supply, and light quality were controlled to improve amino acids accumulation. However, said systems are not capable of measuring the increase in amino acids, nor does it disclose any measures of the improvement in amino acids accumulation. Hence, said system also does not allow for a user of the plant factory to accurately influence the level of amino acids in the end product (i.e., the harvested plant).
[0037] Present indoor growing systems do not allow precise monitoring and control of the nutritional value of crops. Thus, while such systems may be said to improve the growing process of the plant, such improvements are mainly beneficial for the owner of the indoor growing environment because the growing process is more efficient, provides for better energy control, or produces more yield. Present systems are not directed at improving the quality in terms of functional food component content, the healthiness of the end product (i.e., the harvested plant) which may benefit the end user, i.e., the consumer of the harvested crop. None of the prior art gives evidence of elucidating the process of generating functional food components in living plantsduring the growth process. Therefore, present systems are not capable of determining or designing a growing “recipe” for crops in an indoor growing environment that is optimized for an end-goal of a user of the growing environment which relates to the nutritional value - in terms of specific desired functional food components - of the crops at harvest.
[0038] The aim of the present invention is to overcome some of the disadvantages identified in these prior art systems, or to at least provide an alternative for these prior art systems.
[0039] Plant Growing System
[0040] According to a first aspect of the invention, a plant growing system is provided, the plant growing system comprising: an indoor growing environment comprising a sensor system and an actuator system and a controller which is in communication with the sensor system and the actuator system. The controller comprises a control algorithm, the control algorithm being configured to perform a plant growing method comprising the steps of obtaining a present sensor value from the sensor system of the indoor growing environment; and controlling the actuator system of the indoor growing environment with the control algorithm.
[0041] The controller further com prises a recipe for achieving a desired functional food component concentration of plants to be grown in the indoor growing environment. The control algorithm, when performed by the controller in the step of controlling the actuator system, causes the controller to generate a control signal for the actuator system in dependence of the present sensor value such that the recipe is followed.
[0042] The plants to be grown in the indoor growing environment are, in use of the indoor growing environment, planted in a substrate of the indoor growing environment, after which these plants are grown in the indoor growing environment throughout their growth cycle - i.e. , up to the harvest of the plants at harvest time. In other words, once planted in the indoor growing environment, the plants to be grown are plants growing in the indoor growing environment. The growth of the plants is controlled by the actuator system and monitored by the sensor system. The recipe comprises information that is usable by the controller to derive control signals, the control signals being usable to control the actuator system of the indoor growing environment throughout the growth cycle. For example, the information comprises in the recipe may be used by the controller to derive an actuator signal path to be followed by the actuator system. The actuator signal path is an evolution of the actuator signal over time - i.e., from the start of the growth cycle (initial time point) to the end of the growth cycle (end time point). The controller controls the actuator system in accordance with the actuator path.
[0043] By controlling the actuator system of the indoor growing environment such that the recipe is followed, the method ensures optimal conditions in order for the functional food componentconcentration of the plants to be grown approaches the desired functional food component concentration. By precisely controlling the environment in which the plants are grown, it is possible to affect the generation of the functional food components in these plants without having to rely on other techniques such as GMO, bio-fortification of the seeds, etc. Instead, the plant itself is allowed to develop optimally, i.e. , the strength which is inherent to the (natural) plant is used and allowed to fully come to fruition by growing in optimal conditions - which are encoded in the recipe.
[0044] The sensor system comprises one or more sensors, and the actuator system comprises one or more actuators. For example, the sensor system is configured to measure at least one environmental condition of the indoor growing environment, and the actuator system is configured to influence the at least one environmental condition. For example, the at least one environmental condition comprises one or more of temperature, humidity, CO2 level, light intensity and spectral quality. Examples of suitable sensors and actuators are given in the table below:
[0045] Environmental Condition Sensor Actuator
[0046] temperature temperature sensor heater, cooler, HVAC-unit humidity humidity sensor (de)humidifier
[0047] CO2 level CO2 sensor HVAC-unit
[0048] light intensity photoelectric devices (LED)-lights
[0049] spectral quality spectroscope (LED)-lights
[0050]
[0051] In certain other examples, the at least one environmental condition comprises one or more of water availability and nutrient availability. For example, in an indoor growing environment which uses a hydroponics system, the plants growing in the indoor growing environment are grown in a liquid containing a water-based nutrient solution, in which their roots are exposed to said liquid. In such examples, the indoor growing environment is configured to control the flow of said liquid throughout the hydroponics system as well as the concentration of the water-based nutrient solution. The recipe may thus comprise information relating to a prescribed water availability and nutrient availability over time. An actuator suitable for controlling said water availability and / or nutrient availability is controlled by the controller in a way that said prescribed water availability and nutrient availability overtime is followed.
[0052] In general, each of the above-described sensors are typically configured to convert a physical quantity representative of the environmental condition into an electrical signal. Said electrical signal represents a measure of the physical quantity and is read by the controller. The measurement of the environmental condition is subsequently used in the control loop of the control algorithm such that the environmental condition approaches the prescribed environmental condition in the recipe.
[0053] Through the control of the actuator system, the growth of the plants growing in the indoor growing environment is influenced. It was found that, by influencing the growth of said plants - forexample, by influencing the growth conditions of the plants in the indoor growing environment -according to a particular recipe during a growth cycle of said plants, the concentration of a target functional food component which is generated in the plants is influenced in a way that the desired functional food component concentration of the plant is achieved. The recipe is optimized to achieve a predetermined goal of a functional food concentration in the plants. This predetermined goal is for example a concentration of the functional food component at a particular time during the growth cycle, more in particular at harvest of the plant. Thus, by controlling the actuator system according to the recipe, the controller precisely influences the growth of the plant in the indoor growing environment in order to achieve the desired - i.e., predetermined - functional food component concentration in the plants growing in the indoor growing environment.
[0054] For example, the functional food component concentration of the plants to be grown is a concentration of one or more of the following functional food components: a vitamin, such as vitamin C; an amino acid; a peptide; a protein; a polyunsaturated fatty acid; a mineral or trace element, such as calcium, magnesium, and potassium; a carotenoid; an antioxidant, such in bioactive components in tea leaves and berries; dietary fibre; a glucosinolate; an isothiocyanate; an enzyme; and a flavour compound. In the below table, examples for each of these functional food components are provided.
[0055] A vitamin Vitamin C
[0056] An Amino Acid Threonine (symbol Thr or T) is an amino acid that is used in the biosynthesis of proteins
[0057] A Peptide ACE (Angiotensin-Converting Enzyme) inhibitory peptide is a peptide that can help lower bloodpressure.
[0058] A protein Peroxidase (= a group of enzymes = a protein)
[0059] A polyunsaturated fatty acid Alpha-Linolenic acid (ALA) is an Omega-3 type of PUFA (PUFA) for heart health
[0060] A mineral or trace element Such as calcium, magnesium, and potassium Carotenoid P-Carotene (pro-vitamin A)
[0061] An antioxidant such in bioactive Vitamin C, catechins
[0062] components in tea leaves and
[0063] berries
[0064] Dietary fiber Pectin, Mucilage
[0065] A glucosinolate Sinigrin
[0066] A Sulforaphane Sulforaphane
[0067] An enzyme Myrosinase
[0068]
[0069] A flavour compound Flavonoids such as Quercetin, Kaempferol and Myricetin
[0070] Amino Acids are the basic building block. 2 or more (>50) Amino Acids form a Peptide (also referred to as short protein). As more Amino Acids join, the chain becomes a polypeptide. A protein is one or more polypeptide chains that folded into a specific shape and are functionally active. A protein = 1 or more polypeptides = long chains of amino acids held together by peptide bonds.
[0071] In some examples, the controller comprises a plurality of recipes. Each of these recipes is for example optimized for a different desired - i.e., predetermined - functional foodcomponent concentration. In such a case, the plant growing system is configurable by a user of the plant growing system by selecting the recipe which corresponds to the user’s desired functional food component concentration in the plants to be grown in the indoor growing environment. The selected recipe is subsequently used by the controller in the control algorithm to generate the control signal(s) for the actuator system. In some examples, the desired functional food component concentration comprises a plurality of desired concentrations of multiple functional food components. In such cases, the recipe followed by the control algorithm is optimized for each of the desired concentrations of the multiple functional food components.
[0072] In an embodiment according to the first aspect of the invention, the recipe is a function of the at least one environmental condition of the indoor growing environment over time, the recipe prescribing a path in time of the at least one environmental condition to achieve the desired functional food component concentration at harvest of the plants to be grown in the indoor growing environment.
[0073] Thus, the recipe is a function Cenv= f(t), wherein f is the path in time. At any time instant t during the growth cycle of the plants growing in the indoor growing environment, the controller uses the function f to determine a prescribed environmental condition. Based on said prescribed environmental condition and the present environmental condition, as measured by the sensor system, the controller generates a control signal for the actuator system in order for the environmental condition in the indoor growing environment to approach the prescribed environmental condition.
[0074] In an embodiment according to the first aspect of the invention, the plant growing system further comprises a database system with one or more recipes based on historical data of reference plants, and / or historical data of reference plants. The plant growing method further comprises the step of obtaining the recipe for achieving the desired functional food component concentration of the plants to be grown from the database system on the basis of the desired functional food component concentration of the plants to be grown and either the one or more recipes of the database system, or the historical data of the database system, or both the one or more recipes of the database system and the historical data of the database system.
[0075] The reference plants are plants that are in some way biosimilar to the plants to be grown. This may for example mean that the reference plants and the plants to be grown belong to the same family, genus, species, or plant variety.
[0076] The database system is in communication with the controller.
[0077] The recipe is obtained from the database system at the start of the growth cycle of the plants to be grown in the indoor growing environment or during the growth cycle when needed.The historical data on which the recipe is based is obtained during one or more growth cycles of the reference plants. These reference plants were for example grown in a comparable indoor growing environment, for example an identical or the same indoor growing environment in which the plants to be grown are growing / are to be grown. Alternatively, the reference plants were for example grown in a similar indoor growing environment, similar meaning for example that the same environmental conditions are measured and influenced during the growth cycle of the plants in both indoor growing environments. The reference plants were optionally grown in an environmentally controlled manner.
[0078] In any case, the historical data is representative of growth data obtained during the growth cycle of the reference plants. For example, the historical data comprises sensor data representative for data obtainable from the sensor system of the indoor growing environment and obtained during one or more growth cycles of the reference plants; and / or functional food component concentrations of the reference plants. In this example, the same or similar sensors were used to obtain said growth data of the reference plants as the sensors used to obtain growth data of the plants growing in the indoor growing environment. This ensures that the results obtained from the reference plants are transposable to the plants to be grown in the indoor growing environment. As a further example, the historical data comprises functional food component concentrations of the reference plants determined at harvest of the reference plants, the functional food component concentrations at harvest of the reference plants being associated with the sensor data obtained during the one or more growth cycles of the reference plants. As a final example, the historical data comprises functional food component concentration of the reference plants, the concentrations being determined at a plurality of distinct time points during the one or more growth cycles of the reference plants.
[0079] In general, the historical data stored in the database system comprises information (i.e., growth data) pertaining to one or more growth cycles of reference plants which in some way are similar to the plants to be grown or growing in the indoor growing environment - e.g., the reference plants are of the same species as the plants to be grown or growing in the indoor growing environment, or in some other way the reference plants and the plants to be grown or growing in the indoor growing environment share a biological trait.
[0080] By monitoring the growth of the reference plants and saving the growth data - e.g., sensor data representative of environmental conditions and / or functional food component concentrations of the reference plants - it becomes possible to make inferences about that growth data in order to design recipes to achieve a certain predetermined goal. For example, in the growth data, it may be observed that a certain evolution of an environmental condition as measured by one or more sensors in the indoor growing environment has an effect on the generation of, and ultimately the concentration of, a functional food component in the reference plants. By collecting said data throughout numerous growth cycles of the reference plants, a big data set is obtained which isstructured in a way that sensor data and functional food component concentrations of the reference plants at different times during a growth cycle are related to each other and also to that particular growth cycle. In other words, an evolution of one or more environmental conditions based on the sensor data is obtained which is associated with the functional food component concentrations of the reference plants. For example, the functional food component concentration of the reference plants is obtained at harvest of the reference plants. In such an example, the evolution of the environmental conditions is associated to a particular outcome, namely the resulting functional food component concentration at harvest of the reference plants. In another example, the functional food component concentration of the reference plants is obtained throughout the growth cycle of the reference plants, resulting in a functional food component concentration evolution. Said functional food component concentration evolution is associated with the evolution of the environmental conditions. In this way, a more fine-grained inference is made about the influence of the environmental conditions during the growth cycle on the evolution of the functional food component concentration.
[0081] Said inferences are used to create the recipes - which are optionally also stored in the database system - for the plants to be grown in the indoor growing environment. For example, machine learning techniques are used to relate, on the basis of the historical data obtained throughout many growth cycles of the reference plants, an evolution of the environmental conditions in the historical data to a resulting functional food component concentration. Said resulting functional food component concentration is then also the to be expected functional food component concentration of plants growing in the indoor growing environment when a recipe, which uses said evolution of the environmental conditions, is followed in a future growth cycle of plants to be grown in the indoor growing environment. Therewith, it is possible to precisely control the growth of the plants to be grown in the indoor growing environment to achieve a particular predetermined goal, namely a desired functional food component concentration in these plants -measured at harvest of these plants or at intermediate time points during the growth cycle of these plants. Thus, this learning of data from intermediate time points during the growth cycles makes the recipes dynamic. Therewith, a Digital Dynamic Recipe (DDR®) is obtained.
[0082] In certain cases, the database system already stores a recipe which has a desired outcome, namely a to be expected functional food component concentration which corresponds closely to the desired functional food concentration in the plants to be grown in the indoor growing environment. In such a case, the recipe for the plants to be grown is obtained by a query from the controller to the database system, and in response to this query, the database system returns a recipe of which the to be expected functional food component concentration has a closest match to the desired functional food concentration.
[0083] In a variant of this embodiment, the database system comprises a recipe generator configured to generate the recipe for achieving the desired functional food componentconcentration of the plants to be grown based on the historical data stored in the database system. In this embodiment, the recipe generator is configured to perform the following steps:
[0084] • receiving, from the controller, a request for obtaining the recipe for achieving the desired functional food component concentration of the plants to be grown;
[0085] • generating the recipe for achieving the desired functional food component concentration of the plants to be grown on the basis of the historical data stored in the database system;
[0086] • returning, to the controller, the generated recipe for achieving the desired functional food component concentration of the plants to be grown.
[0087] In this variant, the database system may for example not yet store a recipe of which the to be expected functional food component concentration corresponds closely to the desired functional food concentration in the plants to be grown in the indoor growing environment. In such a case, the recipe generator may generate a recipe, for example using machine learning techniques, based on the historical data which is stored in the database system.
[0088] In this variant, the controller is configured to perform the following steps:
[0089] • transmitting, to the database, the request for obtaining the recipe for achieving the desired functional food component concentration of the plants to be grown;
[0090] • receiving, from the database, the generated recipe for achieving the desired functional food component concentration of the plants to be grown.
[0091] In an embodiment according to the first aspect of the invention, the indoor growing environment further comprises a functional food component concentration sensor configured to measure a quantity representative of the functional food component concentration of the plants to be grown, wherein the controller is in communication with the functional food component concentration sensor. In this embodiment, the controller further comprises a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the plants to be grown. In this embodiment, the plant growing method further comprises the following steps:
[0092] • obtaining a present functional food component concentration of the plants to be grown on the basis of a measurement of the functional food component concentration sensor;
[0093] • obtaining, from the functional food component evolution model, a prediction of the functional food component concentration of the plants to be grown;
[0094] • determining a difference between the predicted functional food component concentration and the desired functional food component concentration;
[0095] • in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference; and / or• in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference.
[0096] The functional food component evolution model is a growth model of the functional food component concentration of plants - such as e.g. the reference plants and / or the plants growing in the indoor growing environment - over time. Said growth model is obtained on the basis of the historical data stored in the database system, in particular on the basis of the functional food component concentration measurements of the reference plants. The growth model thus represents a to-be-expected path of the growth of the functional food component concentration over time. The growth model may be parameterizable, for example wherein the parameters relate to the environmental conditions of the indoor growing environment. Thus, a growth model for a particular indoor growing environment is obtainable by providing the necessary parameters to the growth model.
[0097] For example, the database system comprises the functional food component evolution model. The controller obtains the functional food component evolution model from the database system by querying the database system.
[0098] During a growth cycle of the plants growing in the indoor growing environment, a functional food component concentration of the plants is measured. Said measured functional food component concentration is provided to the functional food component evolution model, which infers a to-be-expected (i.e., predicted) functional food component concentration of the plants growing. In case that the to-be-expected functional food component concentration is too low or too high (as compared to the desired functional foo component concentration), the recipe is adapted in order to reduce the difference between the predicted functional food component concentration and the desired functional food component concentration. Said adaptation of the recipe may also be called an intervention, wherein the intervention is an action performed by the controller in the growth cycle of the plants growing which deviates from the recipe. The intervention is for example obtained by the controller querying the database system, the database system providing the intervention based on the difference and the historical data of the database system. The intervention may for example adapt or modify the at least one environmental condition prescribed by the recipe. The prescribed environmental condition may comprise one or more of temperature, humidity, CO2 level, light intensity and spectral quality. Thus, the intervention may change how the at least one environmental condition is influenced by the recipe at an intervention point. The intervention may be a single action to be performed by an actuator at an intervention point, or may comprise several actions to be performed by one or more actuators for modifying / adapting one or more environmental conditions during an intervention period starting at the intervention point.In a variant of this embodiment, the controller is configured to perform the step of adapting the recipe in order to reduce the difference by determining an intervention to be performed and adapting the recipe according to the intervention when the present functional food component concentration of the plants to be grown is above a predefined sweet spot threshold. A sweet spot is a short period during the growth cycle where nutritional functional food components (start to) come to expression. This sweet spot can be detected by taking intermediate measurements during the growth cycle, such as with a hyperspectral camera. Comparative analysis of the values of hyperspectral curves in intermediate measurements - which are representative of the functional food component concentration of the plant - indicate the appearance of the sweet spot. In other words, the sweet spot is an anchor point at which the prediction of the functional food component concentration can be optimized. When entering the sweet spot (i.e., when the present functional food component concentration is above a predefined sweet spot threshold because it crosses or has crossed the predefined sweet spot threshold), functional food components can be accurately determined. And therefore, the recipe can be adapted at the moment the sweet spot is entered, because at that point in time during the growth cycle, an accurate prediction of the resulting functional food component concentration is performed. Said adaptation optimizes the functional food component generation such that the predicted functional food component concentration approaches the desired functional food component concentration. The sweet spot threshold is a value which represents that the functional food component concentration is measurable because the plant has reached a point in the growth cycle in which the functional food component comes to expression. Thus, the sweet spot threshold represents that the functional food component concentration is measurable using the functional food component concentration sensor.
[0099] The predefined sweet spot threshold may for example be determined based on historical growth data obtained during one or more growth cycles of one or more (reference) plants. The sweet spot threshold may for example be recorded as a point in time during the growth cycle of a particular plant (variety) and / or as a sensor value obtainable from the functional food component sensor which indicates the appearance of the sweet spot.
[0100] In certain examples, the plants are grown according to a preset recipe (a default recipe, for example wherein the growth conditions are set to a fixed value) before the sweet spot is detected. In other words, the controller is configured to adapt the recipe according to the intervention only when the present functional food component concentration is above the predefined sweet spot threshold. In such examples, after the sweet spot is detected, a recipe for achieving a desired functional food component concentration in the plants growing in the indoor growing environment is obtained, for example from a database. In those examples, the sweet spot marks the starting point for influencing and / or optimizing the functional food component concentration in the plants. Thus, only from the sweet spot onwards, the recipe is adapted toreduce the difference between the predicted functional food component concentration and the desired functional food component concentration when the present functional food component concentration is above the predefined sweet spot threshold.
[0101] Optionally, in this embodiment and / or in its variant(s), the controller is configured to perform the step of obtaining the prediction of the functional food component concentration of the plants to be grown and the step of determining the difference between the predicted functional food component concentration and the desired functional food component concentration (optionally only) when the present functional food component concentration of the plants to be grown is above the predefined sweet spot threshold.
[0102] Optionally, in this embodiment and / or in its variant(s), the plant growing method further comprises the step of detecting that a sweet spot is entered at a point in the growth cycle when the present functional food component concentration of the plants to be grown passes the sweet spot threshold, and the controller is configured to perform the step of adapting the recipe in response to entering the sweet spot. In other words, the recipe is (optionally only) adapted once the sweet spot is entered (or: in response to entering the sweet spot), i.e. , the measured functional food component concentration of the plants to be grown has (recently) passed the sweet spot threshold. Once the sweet spot threshold has been passed, the sweet spot, being the short period during the growth cycle where nutritional functional food components (start to) come to expression is entered and an accurate prediction of the functional food component concentration (at a later point in the growth cycle and / or at harvest) can be performed.
[0103] The entering of the sweet spot may be detected “live”, e.g., by comparing the measured functional food component concentration continuously or periodically with the sweet spot threshold. Alternatively, the sweet spot may correspond to a point in time during the growth cycle of the plants to be grown which is determined by observing the growth of reference plants during previous growth cycles. Thus, the sweet spot in this case may be referred to as an estimate of the point in time at which the functional food component concentration of the plants to be grown will pass the sweet spot threshold.
[0104] Optionally, in this embodiment and / or in its variant(s), the plant growing method further comprises the steps of obtaining, for each of a set of candidate intervention points, a postintervention prediction of the functional food component concentration of the plants to be grown in case the determined intervention is performed at the intervention point; and determining, based on the obtained post-intervention predictions of the functional food component concentration of the plants to be grown, an optimal intervention point from the set of candidate intervention points for performing the determined intervention. The controller is configured to perform the step of adapting the recipe in order to reduce the difference such that, at the optimal intervention point, the intervention is performed. The optimal intervention point occurs after the present functional food component concentration of the plants to be grown has passed thepredefined sweet spot threshold, i.e., the optimal intervention point occurs after the sweet spot has been entered. The optimal intervention point may for example correspond to the intervention point associated with the highest post-intervention prediction of the functional food component concentration of the plants to be grown. In another example, the optimal intervention point may correspond to the intervention point associated with the post-intervention prediction of the functional food component concentration of the plants to be grown which is closest to the desired functional food component concentration of the plants to be grown. The post-intervention predictions of the functional food component represent a prediction of the functional food component concentration in the plants to be grown if the intervention is performed at the respective intervention point. Once the optimal intervention point is determined, the recipe is adapted: at the intervention point (which corresponds to a (set of) action(s) yet to be performed in the recipe), the planned actions of the recipe are adapted according to the intervention.
[0105] Optionally, in this embodiment and / or in its variant(s), the step of adapting the recipe in order to reduce the difference further comprises the step of requesting, from the database system, an adapted recipe for achieving the desired functional food component concentration of the plants to be grown on the basis of the difference between the predicted functional food component concentration and the desired functional food component concentration.
[0106] For example, the functional food component concentration sensor is a spectral sensor (for example a hyper- or multi-spectral sensor such as a hyper- or multi-spectral camera) which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment. Raw spectral data is obtained from the spectral sensor. This raw spectral data is pre-processed and the resulting pre-processed spectral data is then used to infer the functional food component concentration of the plants. It was found that there is a high correlation between the spectral data and the functional food component concentration of the plants.
[0107] The database system may comprise, as part of the historical data of the reference plants, spectral data which was obtained during the growth cycles of the reference plants. For example, the spectral data comprises hyperspectral images of parts of the plants such as the leaves of the plants. The hyperspectral images usually comprise 200+ channels covering a continuous range of wavelengths e.g. 400-1000nm, usually referred to as the VIS-NIR range (Visible - Near Infrared). A hyperspectral image is for example a 224-channel image of size 1024x1882, stored as a tensor (or multidimensional array) of size 1024x1882x224. Additionally or alternatively, the spectral data comprises multispectral images of parts of the plants such as the leaves of the plants. The multispectral images usually comprise from 4-10 channels corresponding to specific wavelengths. The spectra data may further comprise spectral curves which are derived from the hyper- or multi-spectral images. These hyperspectral curves comprise for each of the wavelengths a measurement of the spectral reflectance of the plants.The database system may further comprise, as part of the historical data of the reference plants, tested functional food component concentrations which are obtained by, for example, destructive testing of the reference plants using known methods.
[0108] The spectral sensor is for example calibrated based on the spectral data which was obtained during the growth cycles of the reference plants and the tested functional food component concentrations of the reference plants. Said calibration may comprise deriving a functional food component concentration model which correlates spectral data representative of the spectral reflectance of the reference plants with functional food component concentrations of said reference plants. The functional food component concentration calculation model is for example obtained using a statistical method such as Partial Least Squares Regression, PLSR, or a deep learning method such as Neural Networks (NN) to correlate, in the historical data, the spectral data representative of the spectral reflectance of the reference plants and the tested functional food component concentrations of the reference plants.
[0109] In an embodiment according to the first aspect of the invention in which the sensor system is configured to measure at least one environmental condition of the indoor growing environment, the plant growing method further comprises the following steps:
[0110] • obtaining, from the sensor system, at harvest of the plants to be grown and optionally at multiple other moments in time during a growth cycle of the plants to be grown, a set of growth data comprising data representative of a measurement of the at least one environmental condition, which measurement is associated to the moment in time that the measurement is performed, the moment in time being relative to a growth cycle of the plants to be grown;
[0111] • storing, in the database system, the set of growth data to be included in the historical data.
[0112] In addition, the growth data may also comprise a measurement of the functional food component concentration of the plants growing in the indoor growing environment, which measurement is associated to the moment in time that the measurement is performed.
[0113] In this embodiment, growth data obtained during the growth cycle of the plants growing in the indoor growth environment are also stored in the database system, alongside the growth data obtained during the growth cycle(s) of the reference plants. This additional data stored in the database system further allows for the recipes stored in the database system and / or generated by the recipe generator to be refined.
[0114] In an embodiment according to the first aspect of the invention comprising a database system with historical data as described above, the one or more recipes are based on historical data stored in the database system obtained by performing the method according to the second aspect of the invention (as described below), and / or the historical data stored in the databasesystem is obtained by performing the method according to the second aspect of the invention (as described below). Similarly, in case the database system comprises a recipe generator, the recipe generator may generate recipes based on historical data stored in the database system obtained by performing the method according to the second aspect of the invention (as described below).
[0115] The plant growing system according to the first aspect of the invention is for example used for two purposes: 1) to grow plants according to established recipes which are used to achieve a desired functional food component concentration in these plants; and 2) to establish said recipes by experimenting with the growth of plants in the indoor growing environment. The first use is subsequently referred to as the “use phase” and the second use is subsequently referred to as the “research phase”. These phases are explained in detail below.
[0116] Research Phase
[0117] According to a second aspect of the invention, a method of growing plants is provided comprising the following steps:
[0118] • growing reference plants indoor, optionally in an environmentally controlled manner, whilst gathering, at harvest of the reference plants and optionally at multiple other moments in time during a growth cycle of the reference plants, a set of growth data, and wherein each set of growth data comprises data representative of a measurement of:
[0119] - a functional food component concentration of the reference plants, and
[0120] - the optionally controlled at least one environmental condition;
[0121] which measurements are associated to the moment in time that the measurements are performed, the moment in time being relative to a growth cycle of the reference plants from which the data are gathered;
[0122] • storing the sets of growth data as historical data in a database.
[0123] The growth data is for example multi-modal historical data which comprises telemetry data of measured environmental conditions such as temperature, humidity, pH, etc.
[0124] The set of growth data is a combination of two measurements at a specific point during the growth cycle of the reference plants. In one example, that specific point is at harvest of the plants. A set comprising the measurement of the at least one environmental condition and the functional food component concentration of the reference plants at harvest is stored in the database. The measurement of the at least one environmental condition may comprise several measurements of the at least one environmental condition during the growth cycle - i.e., defining a path of the at least one environmental condition over time. The resulting set of growth data is thus a representation of a correlation between the measurement of the environmental condition and the measurement of the functional food component concentration of the reference plants.This process may be repeated multiple times for multiple sets of reference plants, resulting in the database comprising, for each of the sets of reference plants, sets of growth data stored as historical data in the database.
[0125] The stored historical data is usable to infer statistics on growth behaviour of the reference plants, the statistics relating to the generated functional food component concentration of the reference plants in relation to the environmental conditions in which these reference plants were grown.
[0126] When growing the reference plants in an environmentally controlled manner, the reference plants are grown in the indoor growing environment by controlling at least one environmental condition of the indoor growing environment during the growth of the reference plants such that a reference recipe for achieving a desired functional food component concentration of the reference plants is followed.
[0127] In an embodiment according to the second aspect of the invention, whilst growing the reference plants indoor, the method further comprises the step of obtaining spectral data from a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment and gathering the measurement of the functional food component concentration of the reference plants based on the obtained spectral data.
[0128] For example, the spectral data comprises hyperspectral images of parts of the plants such as the leaves of the plants. The spectral data may further comprise spectral curves which are derived from the hyper- or multi-spectral images. These hyperspectral curves comprise for each of the wavelengths a measurement of the spectral reflectance of the plants. The database may further comprise, as part of the historical data of the reference plants, tested functional food component concentrations which are obtained by, for example, destructive testing of the reference plants using known methods.
[0129] The spectral sensor is for example calibrated in the same way as explained above for the spectral sensor of the indoor growing environment of the plant growing system according to the first aspect of the invention. In particular, the calibration of the spectral sensor may be performed in accordance with the following process. The historical data in the database comprises spectral data representative of the spectral reflectance of the reference plants. The historical data in the database comprises tested functional food component concentrations of the reference plants and optionally determined functional food component concentration of the reference plants. A functional food component concentration calculation model is obtained by correlating, e.g. using Partial Least Squares Regression, PLSR, in the historical data:
[0130] • the spectral data representative of the spectral reflectance of the reference plants; and • the tested functional food component concentrations of the reference plants.The functional food component concentration calculation model is subsequently used in the step of gathering the measurement of the functional food component concentration of the reference plants by feeding the obtained spectral data - i.e. , a hyper- or multi-spectral image or a hyperspectral curve - into the functional food component concentration calculation model.
[0131] In the step of gathering the set of growth data, the following steps are performed, for example at harvest of the plant and optionally at multiple points in time during the growth cycle of the reference plants:
[0132] • receiving spectral data from the spectral sensor in the indoor growing environment; • determining, e.g. using a destructive testing method, the tested functional food component concentration of the reference plants;
[0133] • optionally, obtaining the determined functional food component concentration of the reference plants by feeding the spectral data into the functional food component concentration calculation model.
[0134] The data so obtained, i.e., the tested functional food component concentrations of the reference plants and the determined functional food component concentration of the reference plants, are also stored as historical data in the database.
[0135] In a variant of this embodiment, the functional food component concentration calculation model is refined by performing the following steps:
[0136] • obtaining a generative Al model trained on the spectral data representative of the spectral reflectance of the reference plants in the historical data;
[0137] • generating a plurality of synthetic spectral datasets by using different combinations of hyperparameters of the calculation generative Al model;
[0138] • for each of the plurality of synthetic spectral datasets, using the statistical method to correlate a union of the spectral data representative of the spectral reflectance of the reference plants in the historical data and the synthetic spectral dataset with the tested functional food component concentrations of the reference plants in the historical data, therewith obtaining a retrained functional food component concentration calculation model;
[0139] • retaining the retrained functional food component calculation model with the highest performance as the functional food component concentration calculation model;
[0140] • including the synthetic spectral dataset with the highest performance in the database to be included in the historical data of the reference plants.
[0141] In an embodiment according to the second aspect of the invention, the reference plants are grown in an environmentally controlled manner in an indoor growing environment and the method further comprises the steps of:
[0142] • obtaining the reference recipe for achieving a desired functional food component concentration of the reference plants on the basis of the historical data in the database; and• controlling at least one environmental condition of the indoor growing environment during the growth of the reference plants such that the reference recipe is followed.
[0143] Thus, in this embodiment, the reference plants are grown in the same or a similar way as explained above for growing the plants to be grown in the indoor growing environment of the plant growing system according to the first aspect of the invention. The indoor growing environment in which the reference plants are grown in this embodiment comprises a sensor-actuator system which is configured to measure and control physical quantities representative of the environmental conditions in the indoor growing environment.
[0144] The recipe for growing the reference plants is obtained for example based on historical data obtained during growth cycles of reference plants. In certain other cases, for example when no relevant historical data is available in the database, a baseline recipe may be obtained. The baseline recipe may be seen as a starting point and is for example designed by a plant expert based on their expert knowledge. The baseline recipe comprises, for the at least one environmental condition of the indoor growing environment, a baseline path in time to be followed when growing the reference plants in the indoor growing environment.
[0145] In a variant of this embodiment, the method further comprises the steps of:
[0146] • obtaining a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the reference plants on the basis of the historical data;
[0147] • whilst growing the reference plants indoor:
[0148] - feeding the measurement of the functional food component concentration of the reference plants into the functional food component evolution model and obtaining a prediction of the functional food component concentration of the reference plants;
[0149] - determining a difference between the predicted functional food component concentration and the desired functional food component concentration of the reference plants;
[0150] - in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the reference plants, adapting the recipe in order to reduce the difference; and / or - in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the reference plants, adapting the recipe in order to reduce the difference.
[0151] The predicted functional food component concentration is representative of the to-be-expected functional food component concentration at a future point in the growth cycle, e.g., at harvest of the plants.
[0152] The functional food component evolution model is - as explained above in the context of the first aspect of the invention - a growth model of the functional food component concentrationof plants - such as e.g. the reference plants and / or the plants growing in the indoor growing environment - over time. Said growth model is obtained on the basis of the historical data stored in the database system, in particular on the basis of the functional food component concentration measurements of the reference plants. The growth model thus represents a to-be-expected path of the growth of the functional food component concentration over time. The growth model may be parameterizable, for example wherein the parameters relate to the environmental conditions of the indoor growing environment. Thus, a growth model for a particular indoor growing environment is obtainable by providing the necessary parameters to the growth model.
[0153] As part of the research phase, i.e., as part of the method according to the second aspect of the invention, the functional food component evolution model may be established. The functional food component evolution model is for example obtained by training a machine learning model on the gathered sets of growth data by correlating the measurements representative of the controlled at least one environmental condition with the measurements representative of the functional food concentration of the reference plants. For example, the database system may establish the functional food component evolution model on the basis of the historical data in the database system which is obtained during the growth of the reference plants, such as by curve fitting the historical data representative of the functional food component concentrations of the reference plants. Additionally or alternatively, the functional food component evolution model is obtained using an evolution model statistical method such as Linear Mixed Effects Model, LMM, to correlate the data representative of the functional food component concentrations of the reference plants in the historical data.
[0154] In a further variant of this embodiment, the method further comprises the steps of:
[0155] • determining, on the basis of the historical data in the database, a sweet spot as a moment in time during the growth cycle of the reference plants in which the functional food component concentration of the reference plants is above a predefined sweet spot threshold;
[0156] • performing the step of adapting the recipe in order to reduce the difference by determining an intervention to be performed (wherein optionally, the intervention is to be performed at the sweet spot) and adapting the recipe according to the intervention. Thus, the step of adapting the recipe in order to reduce the difference is performed by determining, at the sweet spot, an intervention to be performed and adapting the recipe according to the intervention.
[0157] A sweet spot is a short period during the growth cycle where nutritional functional food components (start to) come to expression. This sweet spot can be detected by taking intermediate measurements during the growth cycle, such as with a hyperspectral camera. Comparative analysis of the values of hyperspectal curves in intermediate measurements - which are representative of the functional food component concentration of the plant - indicate the appearance of the sweet spot. In other words, the sweet spot is an anchor point at which theprediction of the functional food component concentration can be optimized. When entering the sweet spot, functional food components can be accurately determined. And therefore, the recipe can be adapted at the moment the sweet spot is entered, because at that point in time during the growth cycle, an accurate prediction of the resulting functional food component concentration is performed. The intervention to be performed may be the result of historic growth data which is representative of the influence of growth conditions on functional food component generation in plants. Thus, the intervention optimizes the functional food component generation such that the predicted functional food component concentration approaches the desired functional food component concentration. The sweet spot threshold is a value which represents that the functional food component concentration is measurable because the plant has reached a point in the growth cycle in which the functional food component comes to expression. Thus, the sweet spot threshold represents that the functional food component concentration is measurable using the functional food component concentration sensor.
[0158] In a further variant of this embodiment, the step of adapting the recipe according to the intervention is performed only when the present functional food component concentration is above the predefined sweet spot threshold. Thus, only from the sweet spot onwards, the recipe may be adapted to reduce the difference between the predicted functional food component concentration and the desired functional food component concentration.
[0159] In a further variant of this embodiment, the method further comprises the steps of obtaining, for each candidate intervention point of a set of candidate intervention points, a postintervention prediction of the functional food component concentration of the plants to be grown in case the determined intervention is performed at the intervention point and determining, based on the obtained post-intervention predictions of the functional food component concentration of the plants to be grown, an optimal intervention point from the set of candidate intervention points for performing the determined intervention. The step of adapting the recipe in order to reduce the difference is performed such that, at the optimal intervention point, the intervention is performed.
[0160] The optimal intervention point occurs after the present functional food component concentration of the plants to be grown has passed the predefined sweet spot threshold, i.e. , the optimal intervention point occurs after the sweet spot has been entered. The optimal intervention point may for example correspond to the intervention point associated with the highest post-intervention prediction of the functional food component concentration of the plants to be grown. In another example, the optimal intervention point may correspond to the intervention point associated with the post-intervention prediction of the functional food component concentration of the plants to be grown which is closest to the desired functional food component concentration of the plants to be grown. The post-intervention predictions of the functional food component represent a prediction of the functional food component if theintervention is performed at the respective intervention point. Once the optimal intervention point is determined, the recipe is adapted: at the intervention point (which corresponds to a (set of) action(s) yet to be performed in the recipe), the planned actions of the recipe are adapted according to the intervention.
[0161] In a further variant of this embodiment, the method further comprises the steps of:
[0162] • obtaining a generative artificial intelligence, Al, model trained on the measurements of the functional food component concentrations in the historical data;
[0163] • generating a plurality of synthetic functional food component concentration datasets by using different combinations of hyperparameters of the generative Al model;
[0164] • for each of the plurality of synthetic functional food component concentration datasets, obtaining a retrained functional food component evolution model on the basis of a union of the functional food component concentrations in the historical data and the synthetic functional food component concentration dataset;
[0165] • retaining the retrained functional food component evolution model with the highest performance as the functional food component evolution model;
[0166] • including the synthetic functional food component concentration dataset with the highest performance in the historical data in the database.
[0167] When referring to the generative Al model used for refining the functional food component concentration calculation model or the functional food component evolution model, the generative Al model is for example a Gaussian Synthesizer, GS, a Conditional Tabular Generative Adversarial Network, CTGAN, or a Tabular Variational Autoencoder, TVAE.
[0168] In an embodiment according to the second aspect of the invention, the method further comprises the steps of:
[0169] • generating a recipe prescribing a path in time of at least one environmental condition to achieve a desired concentration of a functional food component at harvest of the reference plants based on the historical data in the database;
[0170] • storing said recipe in the database.
[0171] Said recipe may then later be used in a next growth cycle of reference plants, and / or in the use phase as explained below.
[0172] Use Phase
[0173] According to a third aspect of the invention, a method of growing plants to be grown in an indoor growing environment is provided,wherein at least one environmental condition of the indoor growing environment is controlled during the growth of the plants to be grown;
[0174] wherein the method comprises the following steps:
[0175] a) obtaining, from a database system, a recipe prescribing a path in time of the controlled at least one environmental condition to achieve a desired concentration of a functional food component at harvest of the plants to be grown;
[0176] b) growing the plants to be grown in the indoor growing environment whilst controlling, in accordance with the recipe, the controlled at least one environmental condition;
[0177] c) harvesting the plants to be grown at harvest time;
[0178] wherein the database system comprises a database of multiple sets of historical data gathered from reference plants, each set being representative of a measurement of:
[0179] • a functional food component concentration of the reference plants, and
[0180] • the controlled at least one environmental condition associated to a moment in time relative to a growth cycle of said reference plants.
[0181] The recipe may for example be the result of the “research phase”, i.e., the method according to the second aspect of the invention as explained above. Based on previous growth cycles of plants that are similar to the plants to be grown, the desired concentration of the functional food component is achieved by following the recipe.
[0182] In an embodiment according to the third aspect of the invention, the database further comprises one or more of said recipes created on the basis of said historical data.
[0183] The recipe creation is further detailed in the “research phase”, i.e., the method according to the second aspect of the invention as explained above.
[0184] In an alternative embodiment according to the third aspect of the invention, the database system comprises a recipe generator configured to generate said recipe based on said historical data.
[0185] In this alternative embodiment, the database system may for example not yet store a recipe of which the to be expected functional food component concentration corresponds closely to the desired functional food concentration in the plants to be grown in the indoor growing environment. In such a case, the recipe generator may generate a recipe, for example using machine learning techniques, based on the historical data which is stored in the database system
[0186] In an embodiment according to the third aspect of the invention, the method further comprises the following steps:• obtaining a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the plants to be grown;
[0187] • obtaining a present functional food component concentration of the plants to be grown on the basis of a measurement of a functional food component concentration sensor of the indoor growing environment;
[0188] • obtaining, from the functional food component evolution model, a predicted functional food component concentration at harvest of the plants to be grown;
[0189] • determining a difference between the predicted functional food component concentration and the desired functional food component concentration;
[0190] • in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference; and / or
[0191] • in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference.
[0192] For example, the functional food component concentration sensor is a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment.
[0193] In a variant of this embodiment, the step of adapting the recipe according to the intervention is performed only when the present functional food component concentration is above the predefined sweet spot threshold.
[0194] In a variant of this embodiment, the method further comprises the step of detecting that a sweet spot is entered at a point in the growth cycle when the present functional food component concentration of the plants to be grown passes the sweet spot threshold, and, the step of adapting the recipe is performed in response to entering the sweet spot.
[0195] In a variant of this embodiment, the method further comprises the steps of obtaining, for each candidate intervention point of a set of candidate intervention points, a post-intervention prediction of the functional food component concentration of the plants to be grown in case the determined intervention is performed at the intervention point; and determining, based on the obtained post-intervention predictions of the functional food component concentration of the plants to be grown, an optimal intervention point from the set of candidate intervention points for performing the determined intervention. The step of adapting the recipe in order to reduce the difference is performed such that, at the optimal intervention point, the intervention is performed.
[0196] In a variant of this embodiment, the step of adapting the recipe in order to reduce the difference further comprises the step of requesting, from the database system, an adapted recipefor achieving the desired functional food component concentration of the plants to be grown on the basis of the difference between the predicted functional food component concentration and the desired functional food component concentration.
[0197] Optionally, in this embodiment or its variant(s), the step of adapting the recipe in order to reduce the difference comprises the step of determining an intervention to be performed and adapting the recipe according to the intervention when the present functional food component concentration of the plants to be grown is above a predefined sweet spot threshold.
[0198] In an embodiment according to the third aspect of the invention, the method further comprises the following steps:
[0199] • obtaining, from a sensor system of the indoor growing environment, at harvest of the reference plants and optionally at multiple other moments in time during a growth cycle of the plants to be grown, a set of growth data comprising data representative of a measurement of the at least one environmental condition, which measurement is associated to the moment in time that the measurement is performed, the moment in time being relative to a growth cycle of the plants to be grown;
[0200] • storing, in the database system, the set of growth data to be included in the historical data.
[0201] In an embodiment according to the third aspect of the invention, before the steps a), b), and c), the database is established by performing the method according to the second aspect of the invention.
[0202] In an embodiment according to the third aspect of the invention in which the database system comprises a recipe generator, the recipe generator is configured to generate the recipe for achieving the desired functional food component concentration of the plant based on historical data stored in the database obtained by performing the method according to the second aspect of the invention.
[0203] Computer Program
[0204] According to a fourth aspect of the invention, a computer program is provided comprising instructions, which, when the program is executed by a computer, cause the computer to carry out the sequence of steps of the method according to the second or the third aspect of the invention or the plant growing method of the plant growing system according to the first aspect of the invention.The invention is described below with reference to the figures. These figures serve as examples to illustrate the invention, and will not be construed as limiting the scope of the claims. In the different figures, like features are indicated by the like reference numerals.
[0205] In the figures:
[0206] Figs. 1A and 1B provide a schematic overview of a plant growing system according to the first aspect of the invention.
[0207] Fig. 2 is a flow diagram which provides a schematic overview of the steps of the method according to the second aspect of the invention.
[0208] Fig. 3 is a flow diagram which provides a schematic overview of an iterative process which is performed as part of the method according to the second and third aspects of the invention.
[0209] Fig. 4 is a flow diagram which provides a schematic overview of an iterative process which is performed as part of the method according to the second and third aspects of the invention.
[0210] Fig. 5A is a flow diagram which provides a schematic overview of the calibration of a spectral sensor which is used to determine the functional food component concentration of plants.
[0211] Fig. 5B is a schematic representation of a conversion between a hyperspectral image and a hyperspectral curve - as part of the calibration process of Fig. 5A.
[0212] Fig. 6 is a flow diagram which provides a schematic overview of a refinement process for the functional food component concentration calculation model and the functional food component evolution model.
[0213] Fig. 7 is a schematic representation of a process for adapting, at a sweet spot, a recipe according to an intervention during the growth cycle.
[0214] The figures are explained in the context of a non-limiting example, namely the optimization of vitamin C concentrations in the Brassicaceae (cabbage) family, which comprises for example pak choi, choi sam, kale, tatsoi, amsoi, and kailaan. These plants all belong to species of the Brassica genus.
[0215] Pak choi and tatsoi belong to the B. rapa species, while Amsoi is a B. juncea variety, and choi sam, kale, and kailaan belong to the B. juncea species.
[0216] The plants were grown in an indoor growing environment which supports for the following range in setpoints in environmental conditions:
[0217] Temperature: -10 - +45 degrees Celsius;
[0218] Humidity: 40% - 90% RH
[0219] Light (total): 20 - 600 pmol m2 s-1 PAR (white: 500 pmol max; far-red: 100 pmol max)Figs. 1A and 1 B provide a schematic overview of a plant growing system 100 according to the first aspect of the invention.
[0220] The plant growing system 100 comprises an indoor growing environment 101 comprising a sensor system 102 and an actuator system 103. The plant growing system 100 further comprises a controller 104 which is in communication with the sensor system 102 and the actuator system 103, the controller comprising a control algorithm 105. The controller further comprises a recipe 106 for achieving a desired functional food component concentration of plants to be grown in the indoor growing environment 101 and the control algorithm 105, when performed by the controller 104 in the step of controlling the actuator system 103, causes the controller 104 to generate a control signal for the actuator system 103 in dependence of the present sensor value such that the recipe 106 is followed.
[0221] The indoor growing environment 101 comprises a plant growth area 107 - which comprises a plant growing medium 108 in which plants to be grown 109 are planted.
[0222] The plant growing system 100 further comprises a database system 110 with a database 111 comprising one or more recipes 112 based on historical data 113 of reference plants of a same species as the plants to be grown 109 in the indoor growing environment 101, and / or historical data 113 of reference plants of a same species as the plants to be grown 109 in the indoor growing environment 101.
[0223] The controller 104 is in communication with the database system 110.
[0224] The database system 110 optionally comprises a recipe generator 114 configured to generate the recipe 106 for achieving the desired functional food component concentration of the plants to be grown 109 based on the historical data stored in the database system 110 — i.e., in the database 111. Otherwise, the controller 104 directly obtains the recipe 106 if said recipe 106 is stored in the database 111 among the one or more recipes 112.
[0225] The controller 104 further comprises a functional food component evolution model 115 which models a development of the functional food component concentration in time throughout the growth of the plants to be grown 109. The functional food component evolution model 115 is based on the historical data stored in the database 111 - and may also be stored in said database 111 or in the database system 110.
[0226] The indoor growing environment 101 further comprises a functional food component concentration sensor 116 configured to measure a quantity representative of the functional food component concentration of the plants to be grown 109, wherein the controller is in communication with the functional food component concentration sensor 116. The functional food component concentration sensor 116 is a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown 109 in the indoor growing environment 101.Fig. 2 is a flow diagram which provides a schematic overview of the steps of the method according to the second aspect of the invention.
[0227] In this method - which is also called the “training phase” - growth data of reference plants is gathered while the reference plants are growing. This growth data is also referred to as telemetry data. The telemetry data is obtained from sensors which measure one or more environmental conditions in which the reference plants are grown - for example, the telemetry is obtained from the sensor system 102 of the plant growing system 100 (refer to Fig. 1).
[0228] In step S201, in the experiment relating to the optimization of Vitamin C content, the reference plants are grown in an indoor growing environment 101 - such as the one schematically shown in Figs. 1A and 1B. The reference plants thus correspond to the plants to be grown 109 as shown in Fig. 1A. In other words: the reference plants are grown by using the indoor growing environment 101 as explained with reference to the first aspect of the invention. The environmental conditions (temperature, humidity...) were set to remain constant throughout the experiment in the completely controlled environment - as a baseline recipe. However, the environmental conditions of temperature and humidity deviated from the original settings with the passage of time, due to the microclimate effect that developed during the growth process. Aside from the treatment, the lighting was full spectrum white LED light, which covers all the necessary bandwidths to ensure healthy plant growth. The baseline recipe - which was based on expert knowledge - is detailed in the following table:
[0229] Lights on (h) 14
[0230] Lights off (h) 10
[0231] Temperature lights on (°C) 26
[0232] Temperature lights off (°C) 21
[0233] Relative humidity 70
[0234] CO2(ppm) 400
[0235] Daily light integral (mol nr2d'1) 12.6
[0236] Light intensity (pmol m'2s'1PPFD 250
[0237] R:B ratio 7
[0238] pH 6.0
[0239] EC 1.8
[0240]
[0241] In step S202, growth data (also called telemetry data) is gathered. The telemetry data collected by the sensor system 102 is pre-processed to fit an evenly spaced time domain with a window size of 10 minutes, meaning that every 10 minutes, a new value from each sensor is recorded in the database. The data and measurements are collected at the level of the plant to develop precision in growth recipes. The insights into the microclimate as it develops during the growth cycle are important factors to consider when developing ultraprecision growthrecipes - these insights can be gathered through the use of the sensor system 102 in the indoor growing environment 101.
[0242] The growth data may also comprise data related to the functional food component concentration of the plants growing in the indoor growing environment 101 - as captured by the functional food component concentration sensor 116. This data includes measured functional food component concentrations, and spectral data (such as hyperspectral images and spectral curves resulting from these hyperspectral images) which is used to obtain these functional food component concentrations.
[0243] In step S203, the growth data is optionally filtered and pre-processed. For example, the telemetry data collected by the sensors is pre-processed to fit an evenly spaced time domain with a fixed window size, for example wherein the window size is 10 minutes, which means that every 10 minutes, a new value for the sensor is recorded. The (raw) spectral data is passed through a data processing pipeline - which is further explained with reference to Figs. 5A and 5B.
[0244] The growth data can be subdivided in two parts: 1) the spectral curve matrix (or independent variable) may serve as input of a machine learning model; and 2) the FFC concentrations, which relate each sample to one or more FFC values. This dual-sided dataset, encompassing spectral data and FFC concentrations, may serve as the foundation for the training and validation of machine learning models designed to predict FFC concentrations based on spectral characteristics.
[0245] Optionally, the datasets are further pre-processed before they are used for training a machine learning model. This pre-processing includes techniques such as log transform, mean centering, data filtering (such as moving average filtering and / or Savitzky-Golay filtering), outlier removal, and data compression.
[0246] In step S204, the growth data - including, e.g., the raw spectral data, FFC concentrations determined based on the raw spectral data, and the environmental conditions which were measured during the growing of the reference plants - which was collected (and optionally filtered and / or pre-processed) during growth cycle of the plants is stored in the database.
[0247] Fig. 3 is a flow diagram which provides a schematic overview of an iterative process which is performed as part of the method according to the second and third aspects of the invention.
[0248] In step S301, the recipe 106 for achieving the desired functional food component concentration - which may be a baseline recipe - is followed by the control algorithm 105 which is performed by the controller 104. While the recipe is being followed, growth data is collected - and so, optionally, steps S202-S204 as explained with reference to Fig. 2 are also performed.In step S302, a spectral curve which is collected during growth of the plants is fed into an FFC calculation model which is used to calibrate the functional food component concentration sensor 116. The functional food component concentration calculation model is for example obtained using a statistical method such as Partial Least Squares Regression, PLSR, or a deep learning method such as Neural Networks (NN) to correlate, in the historical data, the spectral data representative of the spectral reflectance of the reference plants and the tested functional food component concentrations of the reference plants
[0249] In step S303, a present functional food component concentration - for example, in the experiment, a present concentration of vitamin C - is obtained from the functional food component concentration sensor 116. The FFC concentration measurement is for example based on the result of feeding the spectral curve into the FFC calculation model (step S302).
[0250] In step S304, the FFC concentration measurement - i.e., the present functional food component concentration - is fed into the FFC evolution model 115, which infers a to-be-expected (i.e., predicted) functional food component concentration of the plants growing.
[0251] In step S305, a functional food component concentration prediction - for example, in the experiment, a future concentration of vitamin C, for example a concentration of vitamin C at harvest - is obtained by feeding the present functional food component concentration into the functional food component evolution model. The functional food component concentration prediction is a forecast of the functional food component concentration of the plants in the indoor growing environment at a later point in time in the plants’ growth cycle.
[0252] In step S306, the difference between the predicted functional food component concentration and the desired functional food component concentration is determined.
[0253] In step S307, it is determined whether the difference between the predicted functional food component concentration and the desired functional food component concentration is large. For example, it is determined whether the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the plants to be grown; or, it is determined whether case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the plants to be grown.
[0254] In step S308, the recipe in order to reduce the difference. This results in a recipe which approaches the desired functional food component concentration.
[0255] Fig. 4 is a schematic overview of an iterative process which is performed as part of the method according to the second and third aspects of the invention.
[0256] Starting at the “experiment” step, several recipes are tested in an experiment, and the resulting functional food component concentrations are determined using the functional food component concentration calculation model. This results in a dataset, which comprisesmeasurements of the environmental conditions throughout the growth cycle of the plants in the indoor growth environment 101, as well as the resulting functional food component concentrations. Then, the data is used to adapt a next iteration of the recipes, which are again tested in a new experiment continuing the cycle.
[0257] Throughout this cyclic process, a Continuous Training (CT) cycle uses a small sample of lab tests to check for data drifts and retrains the functional food component concentration calculation model, if necessary, which ensures precise functional food component concentration predictions over time.
[0258] The measure-predict-adapt cycle as presented in Fig. 4 is also explained above with reference to Fig. 3, in which these steps of the cycle are also depicted.
[0259] Fig. 5A is a schematic overview of the calibration of a spectral sensor which is used to determine the functional food component concentration of plants, while Fig. 5B is a schematic representation of a conversion between a hyperspectral image and a hyperspectral curve - as part of the calibration process of Fig. 5A.
[0260] During the growing cycle of the plants (both “reference plants” and / or “plants to be grown” as used throughout the text above), spectral data is obtained from the plants. This is particularly important in the establishment of the recipe - i.e., in the reference phase. The spectral data, in particular hyperspectral images, are passed through a data processing pipeline comprising the steps shown in Fig. 5A.
[0261] Each hyperspectral image contains multiple leaves that form one or more samples. In the step S501, the pixels exclusively corresponding to the leaves of a sample of the plant are gathered and the remainder of the pixels is discarded. In particular, the leaves of each plant are segmented to exclude the background from the spectral curve calculation. This task is performed by an Image Segmentation model - for example the Segment Anything Model (SAM). In combination with a language-based object-detection method such as GroundingDINO, the language model is prompted with keywords such as “plants” or “leaves”, which prompting results in the creation of segmentation masks that correspond exclusively to the plants. Thus, each hyperspectral image comprises spectral measurements at one or more wavelengths of one or more leaves, forming one or more samples.
[0262] In the step S502, each image is coupled to a unique ID number (i.e., a label). The ID is shown clearly in the hyperspectral images. The process of reading the IDs from the images and labelling the masked samples through is performed automatically, for example using the keras-ocr method for Optical Character Recognition (OCR).
[0263] In the step S503, a spectral curve is obtained for the hyperspectral image of each sample by collapsing said hyperspectral image into a single vector through averaging of the reflectance curves of each pixel - as shown in Fig. 5B. A hyperspectral measurement isprocessed into a spectral curve to serve as input to the functional food component concentration calculation model. Intuitively, these spectral curves represent a summary of the overall hyperspectral reflectance of the plant leaves. A spectral curve results from collapsing a hyperspectral image into a single vector through averaging of the reflectance curves of each pixel. Each spectral curve is associated with a label i.e., the unique ID number which is shown in the hyperspectral image from which the spectral curve is calculated.
[0264] In the step S504, a ground truth FFC concentration is received, for example as the result of a (destructive) lab test which is performed on the sample to determine the tested functional food component concentrations of the samples - for example, in the experiment, the lab test is performed to determine the tested concentration of vitamin C of the samples. This ground truth FFC concentration is matched to the calculated spectral curve which is labelled with the unique ID number of the sample.
[0265] The HSI dataset - comprising hyperspectral curves and ground truth FFC concentrations - can then be stored in the database and used to train machine learning models (e.g., the functional food component calculation model and the functional food component evolution model) which derive functional food component concentrations from measured hyperspectral values of the plants.
[0266] The functional food component concentration calculation model is configured to associate patterns in the spectral curves with the concentration of the functional food component. After the spectral curves for each hyperspectral image are calculated, the spectral curves are mapped to the lab results for the corresponding samples. This dataset is usable to train the functional food component concentration calculation model, for example using Partial Least Squares Regression, PLSR. Several preprocessing steps are optionally performed on the functional food component concentration calculation model:
[0267] • Logarithmic transformation is applied to data that exhibits exponential growth or is heavily skewed. By taking the logarithm of the data values, largely separated values are compressed, whereas values that are closer together are put further apart, leading to a more symmetric and interpretable distribution. This transformation is especially beneficial for linear models and algorithms that assume data normality, or when the data exhibits many outliers.
[0268] • Mean centering involves shifting the data distribution so that it has a mean value of zero. This preprocessing technique is valuable because it helps remove bias from the data, ensuring that the model focuses on the relative differences between data points rather than their absolute values. By centering the data around zero, the optimization of the machine learning models is enhanced, allowing for a faster and better modelling of the underlying patterns and relationships in the data.• Data filtering, in a general sense, plays a pivotal role in improving machine learning model training by enhancing the quality and relevance of the input data. Filtering techniques help in removing extraneous information, reducing data dimensionality, and highlighting critical patterns and relationships. This preprocessing step not only streamlines the learning process but also enables models to focus on essential features, leading to more accurate and efficient training, ultimately resulting in better model performance and generalization to unseen data. During the experiments, two types of filtering were tested:
[0269] o Moving average filtering: a widely employed technique for analyzing timeseries data. It involves calculating rolling averages over a specified window of data points. This process smooths out fluctuations in the data, making it easier for the model to identify underlying trends and patterns. By using moving averages, noisy and erratic data can be transformed into a more stable and interpretable form, facilitating model training that is sensitive to meaningful changes over time.
[0270] o Savitzky-Golay filtering: a powerful method for smoothing noisy data. It operates by fitting a polynomial function to subsets of the data and then using these polynomial functions to estimate smoothed values. This technique is particularly useful in signal processing and timeseries analysis, as it reduces noise and removes high-frequency variations while preserving important features in the data. By applying SavGol filtering, the preprocessing step prepares the data for model training by providing a cleaner and more interpretable input.
[0271] • Outlier removal: by removing outliers, the preprocessing step improves model robustness, allowing the model to focus on the genuine patterns and relationships within the data, ultimately leading to more accurate and trustworthy predictions. • Data compression, which may be integrated into PLSR.
[0272] • Hyperparameter tuning of the PLSR algorithm. This includes filtering the training data using either Moving Average or Savitzky-Golay, as well as determining the number of PLSR components to utilize. Additionally, the filtering techniques also allow for parameter exploration such as the window size that smooths the data in the most optimal way. All these parameters are referred to as hyperparameters and they contribute to achieving the highest performance with the PLSR algorithm. Various algorithms have been proposed for smarter and more efficient hyperparameter tuning. Another option is Bayes Optimization, which creates a probabilistic model for the objective function to maximize or minimize. It examines which hyperparameters would provide the best performance of the objectivefunction. In the experiment, the R coefficient was maximized as it represents the degree of correlation between the predictor variables and the target FFC. This optimization is for example performed using wandb.ai platform which automates hyperparameter search and provides interactive experiment tracking.
[0273] The resulting trained and optimized functional food component concentration calculation model is used to calculate, from a hyperspectral curve obtained by a hyperspectral sensor in the indoor growing environment 101, the functional food component concentration of the plants in the indoor growing environment 101.
[0274] Fig. 6 is a flow diagram which provides a schematic overview of a refinement process for the functional food component concentration calculation model and the functional food component evolution model.
[0275] In step S601, a generative artificial intelligence, Al, model is obtained which is trained on the measurements of the functional food component concentrations in the historical data. For example, the generative Al model is a Gaussian Synthesizer, GS, a Conditional Tabular Generative Adversarial Network, CTGAN, or a Tabular Variational Autoencoder, TVAE.
[0276] In step S602, a plurality of synthetic functional food component concentration datasets are generated by using different combinations of hyperparameters of the generative Al model.
[0277] In step S603, for each of the plurality of synthetic functional food component concentration datasets, a retrained functional food component calculation model, respectively retrained functional food component evolution model is obtained on the basis of a union of the functional food component concentrations in the historical data and the synthetic functional food component concentration dataset.
[0278] In step S603, the retrained functional food component calculation model, respectively retrained functional food component evolution model with the highest performance is retained as the functional food component calculation model, respectively functional food component evolution model. The synthetic functional food component concentration dataset with the highest performance is included in the historical data in the database.
[0279] Fig. 7 is a schematic representation of a process for adapting, at a sweet spot, a recipe according to an intervention during the growth cycle. Fig. 7 in particular shows a process of adapting the growth recipe in high-impact intervention periods during the growth process after the sweet spot, where the plant expresses the functional food components, has been entered.
[0280] In Fig. 7, a growth cycle 701 of a plant is shown. The growth cycle starts at the planting of the seed of the plant 702 - or, in some cases, after germination by planting a seedling - and ends at harvest of the (fully-grown, harvestable) plant 703. During the growth cycle, the plant to be grown goes through a series of growth phases. The growth of the plant is controlled by arecipe 710. The plant may be grown in a plant growing system according to any one of the embodiments described above, wherein the controller is configured to perform a plant growing method such that the recipe 710 is followed. To follow the recipe, one or more actuators of the plant growing system are controlled by the controller according to the recipe. The recipe may be seen as a series of steps 711a-711e which are performed by the controller over time (i.e., during the growth cycle of the plant to be grown). The series of steps as outlined in the recipe may be seen as actions to be taken which influence one or more environmental condition of the plant growing system - for example, controlling light intensity or spectrum, humidity, temperature, etc. The recipe 710 is adapted to grow a plant to be grown such that the plant to be grown has, at harvest, a desired concentration of a functional food component. The recipe may have been learned by growing reference plants and recording the environmental conditions over time. Such recipes for growing plants have been discussed in the embodiments above.
[0281] The controller is configured to receive, from a sensor configured to measure the concentration of the functional food component in the plant which is growing in the plant growing system, at several points in time during growth of the plant, a sensor value which is representative of the concentration of the functional food component in the plant. During an initial period of the growth cycle, such a concentration cannot yet be measured by the sensor because the plant has not yet developed sufficiently. For example, the sensor may be a spectral sensor which performs spectral analysis of the leaves of the plant. To perform an accurate measurement, a minimum amount of leaf matter has to have been developed by the plant. This initial period is delimited by points 702 and 705 in the growth cycle. During the initial period, the recipe is static and is represented by actions 711a, which prescribe the environmental conditions for the initial period from planting the seed (or seedling) up until the point 705 at which the plant has sufficiently developed for the sensor to measure the concentration of the functional food component in the plant.
[0282] Period 704 in the growth cycle 701 is referred to as a “sweet spot” and represents a (short) period during the growth cycle during which the functional food components come to expression in the plant. In other words, during period 704, the plant has generated sufficient amounts of the functional food component such that the functional food concentration of the plant is measurable by the sensor. The period 704 starts at point 705 and is terminated by point 706. Starting point 705 corresponds to a point in time at which the functional food component concentration of the plant becomes detectable. In other words, the functional food component concentration as measured by the sensor passes a sweet spot threshold at point 705. The sweet spot threshold thus corresponds to the plant entering the sweet spot 704. End point 706 may correspond to a point in time at which the functional food component concentration asmeasured by the sensor passes a second sweet spot threshold. End point 706 marks the end of the “initial” phase of functional food component generation in the plant.
[0283] Once the plant has entered the sweet spot 704, the functional food component concentration of the plant is measurable by the functional food component concentration sensor. And, once said functional food component concentration is measurable, a prediction of a future functional food component concentration (for example: at harvest) of the plant can be obtained by feeding a present measurement of the functional food component concentration into a functional food component evolution model. The functional food component evolution model allows to forecast the functional food component concentration of the plant at future points in the growth cycle 701 (such as at harvest 703). The functional food component evolution model is generally obtained by back-casting, over several growth cycles, functional food component concentrations at several points during the growth cycle 701 (such as at harvest 703, but optionally also at intermediate points in the growth cycle 701) to different environmental conditions occurring during the growth cycle 701 (as prescribed in the recipe 710 and as measured by the sensor system of the plant growing system). The predicted functional food component concentration obtained when entering the sweet spot may not align fully with the desired functional food component concentration. In such a case, the recipe is adapted by an intervention performed on the recipe 710.
[0284] The intervention may be performed during a period in the growth cycle 701. Two such periods 708a, 708b are identified in Fig. 7. These periods are identified by performing a simulation of the effects of the intervention at different candidate intervention points. The candidate intervention points may correspond to start points of a period at which an intervention is performed. For each candidate intervention point, a post-intervention prediction of the functional food concentration is obtained, for example from the functional food component evolution model. An optimal intervention point may subsequently be determined, for example wherein the optimal intervention point results in the predicted functional food component concentration approaching the desired functional food component concentration. Two such optimal intervention points 707a and 707b are shown in Fig. 7. These optimal intervention points 707a, 707b correspond to the optimization of a first and a second functional food component, wherein the first and the second functional food component are different. The recipe 710 is adapted, i.e., the corresponding actions 711c, 711d to be taken by the actuators during optimal intervention periods 708a and 708b, respectively, are adapted according to their respective interventions. The interventions for example specify a change in one or more environmental conditions during the periods 708a, 708b. The periods 708a, 708b may also be referred to as “high-impact periods” as they generally correspond to periods in which the generation of the respective functional food component in the plant may be maximally influenced.As required, this document describes detailed embodiments of the present invention. However, it must be understood that the disclosed embodiments serve exclusively as examples, and that the invention may also be implemented in other forms. Therefore, specific constructional aspects which are disclosed herein should not be regarded as restrictive for the invention, but merely as a basis for the claims and as a basis for rendering the invention implementable by the average skilled person. Furthermore, the various terms used in the description should not be interpreted as restrictive but rather as a comprehensive explanation of the invention. The word "a" used herein means one or more than one, unless specified otherwise. The phrase "a plurality of" means two or more than two. The words "comprising" and "having" do not exclude the presence of more elements. Reference figures in the claims should not be interpreted as restrictive of the invention. Particular embodiments need not achieve all objects described. The mere fact that certain technical measures are specified in different dependent claims still allows the possibility that a combination of these technical measures may advantageously be applied.
[0285] Further, this document describes the following clauses of the present invention:
[0286] 1. Plant growing system, comprising:
[0287] • an indoor growing environment comprising a sensor system and an actuator system;
[0288] • a controller which is in communication with the sensor system and the actuator system, which comprises a control algorithm, the control algorithm being configured to perform a plant growing method comprising the following steps:
[0289] • obtaining a present sensor value from the sensor system of the indoor growing environment;
[0290] • controlling the actuator system of the indoor growing environment with the control algorithm;
[0291] wherein:
[0292] • the controller further comprises a recipe for achieving a desired functional food component concentration of plants to be grown in the indoor growing environment; and
[0293] • the control algorithm, when performed by the controller in the step of controlling the actuator system, causes the controller to generate a control signal for the actuator system in dependence of the present sensor value such that the recipe is followed.
[0294] 2. Plant growing system according to clause 1, wherein the sensor system is configured to measure at least one environmental condition of the indoor growing environment, and the actuator system is configured to influence the at least one environmental condition.
[0295] 3. Plant growing system according to clause 2, wherein the recipe is a function of the at least one environmental condition of the indoor growing environment over time, the recipe prescribing a path in time of the at least one environmental condition to achieve the desired functional foodcomponent concentration at harvest of the plants to be grown in the indoor growing environment.
[0296] 4. Plant growing system according to any one of the preceding clauses, further comprising a database system with
[0297] - one or more recipes based on historical data of reference plants,
[0298] and / or
[0299] - historical data of reference plants ,
[0300] wherein the plant growing method further comprises the step of obtaining the recipe for achieving the desired functional food component concentration of the plants to be grown from the database system on the basis of the desired functional food component concentration of the plants to be grown and
[0301] - the one or more recipes of the database system,
[0302] and / or
[0303] -the historical data of the database system.
[0304] 5. Plant growing system according to clause 4, wherein the historical data comprises:
[0305] • sensor data representative for data obtainable from the sensor system of the indoor growing environment and obtained during one or more growth cycles of the reference plants;
[0306] and / or
[0307] • functional food component concentrations of the reference plants.
[0308] 6. Plant growing system according to clause 5, wherein the historical data comprises functional food component concentrations of the reference plants determined at harvest of the reference plants, the functional food component concentrations at harvest of the reference plants being associated with the sensor data obtained during the one or more growth cycles of the reference plants.
[0309] 7. Plant growing system according to any one of clauses 5-6, wherein the historical data comprises functional food component concentration of the reference plants, the concentrations being determined at a plurality of distinct time points during the one or more growth cycles of the reference plants.
[0310] 8. Plant growing system according to any one of clauses 4-7,
[0311] wherein the database system comprises a recipe generator configured to generate the recipe for achieving the desired functional food component concentration of the plants to be grown based on the historical data stored in the database system,
[0312] wherein the recipe generator is configured to perform the following steps:
[0313] • receiving, from the controller, a request for obtaining the recipe for achieving the desired functional food component concentration of the plants to be grown;• generating the recipe for achieving the desired functional food component concentration of the plants to be grown on the basis of the historical data stored in the database system;
[0314] • returning, to the controller, the generated recipe for achieving the desired functional food component concentration of the plants to be grown.
[0315] 9. Plant growing system according to any one of the preceding clauses,
[0316] wherein the indoor growing environment further comprises a functional food component concentration sensor configured to measure a quantity representative of the functional food component concentration of the plants to be grown, wherein the controller is in communication with the functional food component concentration sensor;
[0317] wherein the controller further comprises a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the plants to be grown;
[0318] wherein the plant growing method further comprises the following steps:
[0319] • obtaining a present functional food component concentration of the plants to be grown on the basis of a measurement of the functional food component concentration sensor;
[0320] • obtaining, from the functional food component evolution model, a prediction of the functional food component concentration of the plants to be grown;
[0321] • determining a difference between the predicted functional food component concentration and the desired functional food component concentration;
[0322] • in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference; and / or
[0323] • in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference.
[0324] 10. Plant growing system according to clause 9, wherein the controller is configured to perform the step of adapting the recipe in order to reduce the difference by determining an intervention to be performed and adapting the recipe according to the intervention when the present functional food component concentration of the plants to be grown is above a predefined sweet spot threshold.
[0325] 11. Plant growing system according to any one of clauses 9-10, wherein the step of adapting the recipe in order to reduce the difference further comprises the step of requesting, from the database system, an adapted recipe for achieving the desired functional food component concentration of the plants to be grown on the basis of the difference between the predicted functional food component concentration and the desired functional food component concentration.12. Plant growing system according to any one of clauses 9-11,
[0326] wherein the functional food component concentration sensor is a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment.
[0327] 13. Plant growing system according to any one of clauses 2-12, wherein the plant growing method further comprises the following steps:
[0328] • obtaining, from the sensor system, at harvest of the plants to be grown and optionally at multiple other moments in time during a growth cycle of the plants to be grown, a set of growth data comprising data representative of a measurement of the at least one environmental condition, which measurement is associated to the moment in time that the measurement is performed, the moment in time being relative to a growth cycle of the plants to be grown;
[0329] • storing, in the database system, the set of growth data to be included in the historical data.
[0330] 14. Plant growing system according to any one of clauses 4-13, wherein:
[0331] the one or more recipes are based on historical data stored in the database system obtained by performing the method according to any one of clauses 17-24, and / or
[0332] the historical data stored in the database system is obtained by performing the method according to any one of clauses 17-24.
[0333] 15. Plant growing system according to any one of the preceding clauses, wherein the functional food component concentration of the plants to be grown is a concentration of one or more of the following functional food components: a vitamin, such as vitamin C; an amino acid; a peptide; a protein; a polyunsaturated fatty acid; a mineral or trace element, such as calcium, magnesium, and potassium; a carotenoid; an antioxidant, such in bioactive components in tea leaves and berries; dietary fibre; a glucosinolate; an isothiocyanate; an enzyme; a flavour compound.
[0334] 16. Plant growing system according to any one of clauses 2-15, wherein the at least one environmental condition comprises one or more of temperature, humidity, CO2 level, light intensity and spectral quality.
[0335] 17. Method of growing plants, comprising the following steps:
[0336] • growing reference plants in an environmentally controlled manner in an indoor growing environment by controlling at least one environmental condition of the indoor growing environment during the growth of the reference plants such that a reference recipe for achieving a desired functional food component concentration of the reference plants is followed, whilst gathering, at harvest of the reference plants and optionally at multiple other moments in time during a growth cycle of the reference plants, a set of growth data, and wherein each set of growth data comprises data representative of a measurement of:
[0337] - a functional food component concentration of the reference plants, and- the controlled at least one environmental condition;
[0338] which measurements are associated to the moment in time that the measurements are performed, the moment in time being relative to a growth cycle of the reference plants from which the data are gathered;
[0339] • storing the sets of growth data as historical data in a database.
[0340] 18. Method according to clause 17, wherein, whilst growing the reference plants indoor, the method further comprises the step of obtaining spectral data from a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment and gathering the measurement of the functional food component concentration of the reference plants based on the obtained spectral data.
[0341] 19. Method according to any one of clauses 17-18,
[0342] wherein the method further comprises the step of obtaining the reference recipe on the basis of the historical data in the database.
[0343] 20. Method according to clause 19, further comprising the steps of:
[0344] • obtaining a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the reference plants on the basis of the historical data;
[0345] • whilst growing the reference plants indoor:
[0346] - feeding the measurement of the functional food component concentration of the reference plants into the functional food component evolution model and obtaining a prediction of the functional food component concentration of the reference plants;
[0347] - determining a difference between the predicted functional food component concentration and the desired functional food component concentration of the reference plants;
[0348] - in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the reference plants, adapting the recipe in order to reduce the difference; and / or
[0349] - in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the reference plants, adapting the recipe in order to reduce the difference.
[0350] 21. Method according to clause 20, further comprising the steps of:
[0351] • determining, on the basis of the historical data in the database, a sweet spot as a moment in time during the growth cycle of the reference plants in which the functional food component concentration of the reference plants is above a predefined sweet spot threshold;• performing the step of adapting the recipe in order to reduce the difference by determining an intervention to be performed at the sweet spot and adapting the recipe according to the intervention.
[0352] 22. Method according to any one of clauses 20-21 , further comprising the steps of:
[0353] • obtaining a generative artificial intelligence, Al, model trained on the measurements of the functional food component concentrations in the historical data;
[0354] • generating a plurality of synthetic functional food component concentration datasets by using different combinations of hyperparameters of the generative Al model;
[0355] • for each of the plurality of synthetic functional food component concentration datasets, obtaining a retrained functional food component evolution model on the basis of a union of the functional food component concentrations in the historical data and the synthetic functional food component concentration dataset;
[0356] • retaining the retrained functional food component evolution model with the highest performance as the functional food component evolution model;
[0357] • including the synthetic functional food component concentration dataset with the highest performance in the historical data in the database.
[0358] 23. Method according to clause 22, wherein the generative Al model is a Gaussian Synthesizer, GS, a Conditional Tabular Generative Adversarial Network, CTGAN, ora Tabular Variational Autoencoder, TVAE.
[0359] 24. Method according to any one of clauses 17-23, further comprising the steps of:
[0360] • generating a recipe prescribing a path in time of at least one environmental condition to achieve a desired concentration of a functional food component at harvest of the reference plants based on the historical data in the database;
[0361] • storing said recipe in the database.
[0362] 25. Method of growing plants to be grown in an indoor growing environment, wherein at least one environmental condition of the indoor growing environment is controlled during the growth of the plants to be grown;
[0363] wherein the method comprises the following steps:
[0364] a) obtaining, from a database system, a recipe prescribing a path in time of the controlled at least one environmental condition to achieve a desired concentration of a functional food component at harvest of the plants to be grown;
[0365] b) growing the plants to be grown in the indoor growing environment whilst controlling, in accordance with the recipe, the controlled at least one environmental condition;
[0366] c) harvesting the plants to be grown at harvest time;
[0367] wherein the database system comprises a database of multiple sets of historical data gathered from reference plants, each set being representative of a measurement of:• a functional food component concentration of the reference plants, and
[0368] • the controlled at least one environmental condition associated to a moment in time relative to a growth cycle of said reference plants.
[0369] 26. Method according to clause 25, wherein the database further comprises one or more of said recipes created on the basis of said historical data.
[0370] 27. Method according to clause 25, wherein the database system comprises a recipe generator configured to generate said recipe based on said historical data.
[0371] 28. Method according to any one of clauses 25-27, further comprising the following steps:
[0372] • obtaining a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the plants to be grown;
[0373] • obtaining a present functional food component concentration of the plants to be grown on the basis of a measurement of a functional food component concentration sensor of the indoor growing environment;
[0374] • obtaining, from the functional food component evolution model, a predicted functional food component concentration at harvest of the plants to be grown;
[0375] • determining a difference between the predicted functional food component concentration and the desired functional food component concentration;
[0376] • in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference; and / or
[0377] • in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference.
[0378] 29. Method according to clause 28, wherein the step of adapting the recipe in order to reduce the difference further comprises the step of requesting, from the database system, an adapted recipe for achieving the desired functional food component concentration of the plants to be grown on the basis of the difference between the predicted functional food component concentration and the desired functional food component concentration.
[0379] 30. Method according to any one of clauses 28-29, wherein the step of adapting the recipe in order to reduce the difference comprises the step of determining an intervention to be performed and adapting the recipe according to the intervention when the present functional food component concentration of the plants to be grown is above a predefined sweet spot threshold.
[0380] 31. Method according to any one of clauses 28-30,wherein the functional food component concentration sensor is a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment.
[0381] 32. Method according to any one of clauses 25-31 , further comprising the following steps:
[0382] • obtaining, from a sensor system of the indoor growing environment, at harvest of the reference plants and optionally at multiple other moments in time during a growth cycle of the plants to be grown, a set of growth data comprising data representative of a measurement of the at least one environmental condition, which measurement is associated to the moment in time that the measurement is performed, the moment in time being relative to a growth cycle of the plants to be grown;
[0383] • storing, in the database system, the set of growth data to be included in the historical data.
[0384] 33. Method according to any one of clauses 25-32, wherein before the steps a), b), and c), the database is established by performing the method according to any one of clauses 16-23.
[0385] 34. Method according to any one of clauses 27-32, wherein the recipe generator is configured to generate the recipe for achieving the desired functional food component concentration of the plant based on historical data stored in the database obtained by performing the method according to any one of clauses 16-23.
[0386] 35. Computer program comprising instructions, which, when the program is executed by a computer, cause the computer to carry out the sequence of steps of the method according to any one of clauses 17-34 or the plant growing method of the plant growing system according to any one of clauses 1-16.
Claims
47CLAIMS1. Plant growing system, comprising:• an indoor growing environment comprising a sensor system and an actuator system;• a controller which is in communication with the sensor system and the actuator system, which comprises a control algorithm, the control algorithm being configured to perform a plant growing method comprising the following steps:• obtaining a present sensor value from the sensor system of the indoor growing environment;• controlling the actuator system of the indoor growing environment with the control algorithm;wherein:• the controller further comprises a recipe for achieving a desired functional food component concentration of plants to be grown in the indoor growing environment; and• the control algorithm, when performed by the controller in the step of controlling the actuator system, causes the controller to generate a control signal for the actuator system in dependence of the present sensor value such that the recipe is followed,wherein the indoor growing environment further comprises a functional food component concentration sensor configured to measure a quantity representative of the functional food component concentration of the plants to be grown, wherein the controller is in communication with the functional food component concentration sensor;wherein the controller further comprises a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the plants to be grown;wherein the plant growing method further comprises the following steps:• obtaining a present functional food component concentration of the plants to be grown on the basis of a measurement of the functional food component concentration sensor;• obtaining, from the functional food component evolution model, a prediction of the functional food component concentration of the plants to be grown;• determining a difference between the predicted functional food component concentration and the desired functional food component concentration;• in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference; and / or• in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference,48wherein the controller is configured to perform the step of adapting the recipe in order to reduce the difference by determining an intervention to be performed and adapting the recipe according to the intervention when the present functional food component concentration of the plants to be grown is above a predefined sweet spot threshold indicating that the functional food component concentration is measurable.
2. Plant growing system according to claim 1 , wherein the controller is configured to perform the step of adapting the recipe according to the intervention only when the present functional food component concentration is above the predefined sweet spot threshold.
3. Plant growing system according to any one of the preceding claims, wherein the plant growing method further comprises the step of detecting that a sweet spot is entered at a point in the growth cycle when the present functional food component concentration of the plants to be grown passes the sweet spot threshold, wherein the controller is configured to perform the step of adapting the recipe in response to entering the sweet spot.
4. Plant growing system according to any one of the preceding claims, wherein the plant growing method further comprises the steps of:• obtaining, for each candidate intervention point of a set of candidate intervention points, a post-intervention prediction of the functional food component concentration of the plants to be grown in case the determined intervention is performed at the intervention point; and• determining, based on the obtained post-intervention predictions of the functional food component concentration of the plants to be grown, an optimal intervention point from the set of candidate intervention points for performing the determined intervention;wherein the controller is configured to perform the step of adapting the recipe in order to reduce the difference such that, at the optimal intervention point, the intervention is performed.
5. Plant growing system according to claim any one of the preceding claims, wherein the sensor system is configured to measure at least one environmental condition of the indoor growing environment, and the actuator system is configured to influence the at least one environmental condition.
6. Plant growing system according to claim 5, wherein the recipe is a function of the at least one environmental condition of the indoor growing environment over time, the recipe prescribing a path in time of the at least one environmental condition to achieve the desired functional food49component concentration at harvest of the plants to be grown in the indoor growing environment.
7. Plant growing system according to any one of the preceding claims, further comprising a database system with- one or more recipes based on historical data of reference plants,and / or- historical data of reference plants ,wherein the plant growing method further comprises the step of obtaining the recipe for achieving the desired functional food component concentration of the plants to be grown from the database system on the basis of the desired functional food component concentration of the plants to be grown and- the one or more recipes of the database system,and / or-the historical data of the database system.
8. Plant growing system according to claim 7, wherein the historical data comprises:• sensor data representative for data obtainable from the sensor system of the indoor growing environment and obtained during one or more growth cycles of the reference plants;and / or• functional food component concentrations of the reference plants.
9. Plant growing system according to claim 8, wherein the historical data comprises functional food component concentrations of the reference plants determined at harvest of the reference plants, the functional food component concentrations at harvest of the reference plants being associated with the sensor data obtained during the one or more growth cycles of the reference plants.
10. Plant growing system according to any one of claims 8-9, wherein the historical data comprises functional food component concentration of the reference plants, the concentrations being determined at a plurality of distinct time points during the one or more growth cycles of the reference plants.
11. Plant growing system according to any one of claims 7-10,wherein the database system comprises a recipe generator configured to generate the recipe for achieving the desired functional food component concentration of the plants to be grown based on the historical data stored in the database system,50wherein the recipe generator is configured to perform the following steps:• receiving, from the controller, a request for obtaining the recipe for achieving the desired functional food component concentration of the plants to be grown;• generating the recipe for achieving the desired functional food component concentration of the plants to be grown on the basis of the historical data stored in the database system;• returning, to the controller, the generated recipe for achieving the desired functional food component concentration of the plants to be grown.
12. Plant growing system according to any one of the preceding claims, wherein the step of adapting the recipe in order to reduce the difference further comprises the step of requesting, from the database system, an adapted recipe for achieving the desired functional food component concentration of the plants to be grown on the basis of the difference between the predicted functional food component concentration and the desired functional food component concentration.
13. Plant growing system according to any one of the preceding claims,wherein the functional food component concentration sensor is a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment.
14. Plant growing system according to any one of claims 5-13, wherein the plant growing method further comprises the following steps:• obtaining, from the sensor system, at harvest of the plants to be grown and optionally at multiple other moments in time during a growth cycle of the plants to be grown, a set of growth data comprising data representative of a measurement of the at least one environmental condition, which measurement is associated to the moment in time that the measurement is performed, the moment in time being relative to a growth cycle of the plants to be grown;• storing, in the database system, the set of growth data to be included in the historical data.
15. Plant growing system according to any one of claims 7-14, wherein:the one or more recipes are based on historical data stored in the database system obtained by performing the method according to any one of claims 18-26, and / orthe historical data stored in the database system is obtained by performing the method according to any one of claims 18-26.
16. Plant growing system according to any one of the preceding claims, wherein the functional food component concentration of the plants to be grown is a concentration of one or more of the following functional food components:• a vitamin, such as vitamin C;• an amino acid;• a peptide;• a protein;• a polyunsaturated fatty acid;• a mineral or trace element, such as calcium, magnesium, and potassium;• a carotenoid;• an antioxidant, such in bioactive components in tea leaves and berries;• dietary fibre;• a glucosinolate;• an isothiocyanate;• an enzyme;• a flavour compound.
17. Plant growing system according to any one of claims 5-16, wherein the at least one environmental condition comprises one or more of temperature, humidity, CO2 level, light intensity and spectral quality.
18. Method of growing plants, comprising the following steps:• growing reference plants in an environmentally controlled manner in an indoor growing environment by controlling at least one environmental condition of the indoor growing environment during the growth of the reference plants such that a reference recipe for achieving a desired functional food component concentration of the reference plants is followed, whilst gathering, at harvest of the reference plants and optionally at multiple other moments in time during a growth cycle of the reference plants, a set of growth data, and wherein each set of growth data comprises data representative of a measurement of:- a functional food component concentration of the reference plants, and - the controlled at least one environmental condition;which measurements are associated to the moment in time that the measurements are performed, the moment in time being relative to a growth cycle of the reference plants from which the data are gathered;• storing the sets of growth data as historical data in a database,the method further comprising the steps of:• obtaining a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the reference plants on the basis of the historical data;• whilst growing the reference plants in the indoor growing environment:- feeding the measurement of the functional food component concentration of the reference plants into the functional food component evolution model and obtaining a prediction of the functional food component concentration of the reference plants;- determining a difference between the predicted functional food component concentration and the desired functional food component concentration of the reference plants;- in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the reference plants, adapting the recipe in order to reduce the difference; and / or- in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the reference plants, adapting the recipe in order to reduce the difference,• determining, on the basis of the historical data in the database, a sweet spot as a moment in time during the growth cycle of the reference plants in which the functional food component concentration of the reference plants is above a predefined sweet spot threshold indicating that the functional food component is measurable;• wherein the step of adapting the recipe in order to reduce the difference is performed by determining, at the sweet spot, an intervention to be performed and adapting the recipe according to the intervention.
19. Method according to claim 18, wherein the functional food component evolution model is obtained by training a machine learning model on the gathered sets of growth data by correlating the measurements representative of the controlled at least one environmental condition with the measurements representative of the functional food concentration of the reference plants.
20. Method according to any one of claims 18-19, wherein the step of adapting the recipe according to the intervention is performed only when the present functional food component concentration is above the predefined sweet spot threshold.
21. Method according to any one of claims 18-20, further comprising the steps of:53• obtaining, for each candidate intervention point of a set of candidate intervention points, a post-intervention prediction of the functional food component concentration of the plants to be grown in case the determined intervention is performed at the intervention point; and• determining, based on the obtained post-intervention predictions of the functional food component concentration of the plants to be grown, an optimal intervention point from the set of candidate intervention points for performing the determined intervention;wherein the step of adapting the recipe in order to reduce the difference is performed such that, at the optimal intervention point, the intervention is performed.
22. Method according to any one of claims 18-21, wherein, whilst growing the reference plants in the indoor growing environment, the method further comprises the step of obtaining spectral data from a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment and gathering the measurement of the functional food component concentration of the reference plants based on the obtained spectral data.
23. Method according to any one of claims 18-22,wherein the method further comprises the step of obtaining the reference recipe on the basis of the historical data in the database.
24. Method according to any one of claims 18-23, further comprising the steps of:• obtaining a generative artificial intelligence, Al, model trained on the measurements of the functional food component concentrations in the historical data;• generating a plurality of synthetic functional food component concentration datasets by using different combinations of hyperparameters of the generative Al model;• for each of the plurality of synthetic functional food component concentration datasets, obtaining a retrained functional food component evolution model on the basis of a union of the functional food component concentrations in the historical data and the synthetic functional food component concentration dataset;• retaining the retrained functional food component evolution model with the highest performance as the functional food component evolution model;• including the synthetic functional food component concentration dataset with the highest performance in the historical data in the database.5425. Method according to claim 24, wherein the generative Al model is a Gaussian Synthesizer, GS, a Conditional Tabular Generative Adversarial Network, CTGAN, ora Tabular Variational Autoencoder, TVAE.
26. Method according to any one of claims 18-25, further comprising the steps of:• generating a recipe prescribing a path in time of at least one environmental condition to achieve a desired concentration of a functional food component at harvest of the reference plants based on the historical data in the database;• storing said recipe in the database.
27. Method of growing plants to be grown in an indoor growing environment, wherein at least one environmental condition of the indoor growing environment is controlled during the growth of the plants to be grown;wherein the method comprises the following steps:a) obtaining, from a database system, a recipe prescribing a path in time of the controlled at least one environmental condition to achieve a desired concentration of a functional food component at harvest of the plants to be grown;b) growing the plants to be grown in the indoor growing environment whilst controlling, in accordance with the recipe, the controlled at least one environmental condition;c) harvesting the plants to be grown at harvest time;wherein the database system comprises a database of multiple sets of historical data gathered from reference plants, each set being representative of a measurement of:• a functional food component concentration of the reference plants, and• the controlled at least one environmental condition associated to a moment in time relative to a growth cycle of said reference plants,the method further comprising the following steps:• obtaining a functional food component evolution model which models a development of the functional food component concentration in time throughout the growth of the plants to be grown;• obtaining a present functional food component concentration of the plants to be grown on the basis of a measurement of a functional food component concentration sensor of the indoor growing environment;• obtaining, from the functional food component evolution model, a predicted functional food component concentration at harvest of the plants to be grown;• determining a difference between the predicted functional food component concentration and the desired functional food component concentration;55• in case the predicted functional food component concentration is a first predetermined amount smaller than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference; and / or• in case the predicted functional food component concentration is a second predetermined amount larger than the desired functional food component concentration of the plants to be grown, adapting the recipe in order to reduce the difference,wherein the step of adapting the recipe in order to reduce the difference comprises the step of determining an intervention to be performed and adapting the recipe according to the intervention when the present functional food component concentration of the plants to be grown is above a predefined sweet spot threshold indicating that the functional food component is measurable.
28. Method according to claim 27, wherein the step of adapting the recipe according to the intervention is performed only when the present functional food component concentration is above the predefined sweet spot threshold.
29. Method according to any one of claims 27-28, further comprising the step of detecting that a sweet spot is entered at a point in the growth cycle when the present functional food component concentration of the plants to be grown passes the sweet spot threshold, and, wherein the step of adapting the recipe is performed in response to entering the sweet spot.
30. Method according to any one of claims 27-29, further comprising the steps of:• obtaining, for each candidate intervention point of a set of candidate intervention points, a post-intervention prediction of the functional food component concentration of the plants to be grown in case the determined intervention is performed at the intervention point; and• determining, based on the obtained post-intervention predictions of the functional food component concentration of the plants to be grown, an optimal intervention point from the set of candidate intervention points for performing the determined intervention;wherein the step of adapting the recipe in order to reduce the difference is performed such that, at the optimal intervention point, the intervention is performed.
31. Method according to any one of claims 27-30, wherein the database further comprises one or more of said recipes created on the basis of said historical data.
32. Method according to any one of claims 27-30, wherein the database system comprises a recipe generator configured to generate said recipe based on said historical data.5633. Method according to any one of claims 27-32, wherein the step of adapting the recipe in order to reduce the difference further comprises the step of requesting, from the database system, an adapted recipe for achieving the desired functional food component concentration of the plants to be grown on the basis of the difference between the predicted functional food component concentration and the desired functional food component concentration.
34. Method according to any one of claims 27-33,wherein the functional food component concentration sensor is a spectral sensor which is configured to measure the spectral reflectance of the plants to be grown in the indoor growing environment.
35. Method according to any one of claims 27-34, further comprising the following steps:• obtaining, from a sensor system of the indoor growing environment, at harvest of the reference plants and optionally at multiple other moments in time during a growth cycle of the plants to be grown, a set of growth data comprising data representative of a measurement of the at least one environmental condition, which measurement is associated to the moment in time that the measurement is performed, the moment in time being relative to a growth cycle of the plants to be grown;• storing, in the database system, the set of growth data to be included in the historical data.
36. Method according to any one of claims 27-35, wherein before the steps a), b), and c), the database is established by performing the method according to any one of claims 17-25.
37. Method according to any one of claims 32-36, wherein the recipe generator is configured to generate the recipe for achieving the desired functional food component concentration of the plant based on historical data stored in the database obtained by performing the method according to any one of claims 17-25.
38. Computer program comprising instructions, which, when the program is executed by a computer, cause the computer to carry out the sequence of steps of the method according to any one of claims 18-37 or the plant growing method of the plant growing system according to any one of claims 1-17.
