Method for mushroom cultivation

Abstract

The current invention relates to an improved method for mushroom cultivation in one or more stages of the growing process, along with associated systems therefor.

Description

[0001] METHOD FOR MUSHROOM CULTIVATION

[0002] FIELD OF THE INVENTION

[0003] The present invention relates to an improved method for mushroom cultivation, based on machine vision technology. The invention is aimed at leveraging machine vision for an improved interpretation of the growing mycelium, pin s or nodules and mushrooms, and taking necessary actions based on the detected information, along with correlating the detections with other data relevant to the cultivation process.

[0004] BACKGROUND

[0005] Mushrooms are deeply favored by consumers due to the characteristics of delicious taste, rich nutrition, moderate price and the like. Furthermore, due to the possibility to grow mushrooms intensively (i.e., high yield on a relatively limited surface and in various climates), mushroom cultivation has grown more widespread in the past decades, and is expected to rise further in the future.

[0006] However, because of this, a further need arises for optimization and automation, which puts substantial strain on the expert knowledge that is often required in this process, and is difficult to transfer or translate in general rules.

[0007] Rimantas Barauskas et al discuss part of this context in "Approach of Al-based automatic climate control in white button mushroom growing hall" (Agriculture, part 12, nr. 11, 15 November 2022), but fail to resolve one or more of the issues identified in this application.

[0008] In a first aspect, one of the issues in this industry is determination and prediction when harvesting is necessary or even required, both in staggered or mass cultivation, the former usually employing manual harvesting, the latter (also) using automated harvesting. Not only is this knowledge crucial to avoid spoilage, but it also important in terms of planning in production resources, such as manpower.

[0009] In a further aspect, one of the issues in this industry is the identification when other steps in the cultivation process are necessary, such as progressing to a next phase after mycelium growth, initiating primordia formation, pin or nodule formation or nodulation.In a further aspect, and one that builds further on the preceding aspects, an issue is that climate control is crucial in mushroom cultivation, as it influences the optimization very strongly, but also because each phase is very sensitive in terms of progression to a next phase. The thresholds between phases can in some cases easily be triggered, requiring a very careful control of environmental parameters.

[0010] In a further aspect, it is notable that in the process of mushroom planting and production, the picking link of mushrooms is the link with the largest labor consumption and the highest cost at present, and if the mushrooms are not picked timely, the problems of mushroom opening, excessive growth of mushroom bodies, too dense mushroom clusters and the like can be caused, so that the growth of the mushrooms in the next tide is affected, and serious economic loss is caused. As such, further challenges are present, namely in the optimization of the mushroom beds prior to picking by pin or nodule (or pin head) reduction (removal), to allow maximized growth in terms of pickable mushrooms with optimal quality.

[0011] Therefore, there is a need for an efficient method for mushroom removal, that can intelligently identify and locate mushrooms (or pins, pin heads) on a mushroom bed that are suitable for removal or picking.

[0012] In a final aspect, an issue in this industry is that the pin reduction or removal is still performed manually, and requires a workforce that is both well trained (knows which pins to remove) and is careful and capable of removing pins (knows how to remove them). Automation is crucial herein, building further on the previous aspect, but an industrially applicable method or system for performing this pin removal is required. This prevents an excessive number of pins that would interfere with each other's growth by competing for (limited) space and resources, which can result in a reduced yield in terms of quality and quantity.

[0013] SUMMARY OF THE INVENTION

[0014] The present invention and embodiments thereof serve to provide solutions in one or more phases of the process of (edible) mushroom cultivation, not limited to mycelium growth phase, nodulation, "button" growth phase, and formation of the fruiting bodies of the mushrooms themselves. The goal is optimization in quality and quantity of the resulting products in terms of space, investment and time that is occupied per growth cycle. As such, it involves finding the correct time to initiate or trigger each phase, as well as adapting environmental factors (such as climate in thegrowing area during each of the phases, and preferably adapting this based on the detected growth of pins and / or mushrooms) and pin removal / reduction where necessary.

[0015] DESCRIPTION OF FIGURES

[0016] Figure 1A and IB show a perspective view of a monitoring system for performing a method according to an embodiment of the invention, mounted above a mushroom growth bed.

[0017] Figure 2 shows monitored / tracked parameters and environmental parameters over time for a growth bed.

[0018] Figure 3 shows a processed image from a mushroom growth bed, in which individual mushrooms are identified, allowing individual and averaged tracking of parameters.

[0019] Figure 4 shows an exterior perspective of a vertical cultivation system with multiple growth beds.

[0020] Figure 5 shows a detailed view of a level in the cultivation system of Figure 4.

[0021] DETAILED DESCRIPTION OF THE INVENTION

[0022] Unless otherwise defined, all terms used in disclosing the invention, including technical and scientific terms, have the meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. By means of further guidance, term definitions are included to better appreciate the teaching of the present invention.

[0023] As used herein, the following terms have the following meanings:

[0024] "A", "an", and "the" as used herein refers to both singular and plural referents unless the context clearly dictates otherwise. By way of example, "a compartment" refers to one or more than one compartment.

[0025] "About" as used herein referring to a measurable value such as a parameter, an amount, a temporal duration, and the like, is meant to encompass variations of + / -20% or less, preferably + / -10% or less, more preferably + / -5% or less, even morepreferably + / -1% or less, and still more preferably + / -0.1% or less of and from the specified value, in so far such variations are appropriate to perform in the disclosed invention. However, it is to be understood that the value to which the modifier "about" refers is itself also specifically disclosed.

[0026] "Comprise", "comprising", and "comprises" and "comprised of" as used herein are synonymous with "include", "including", "includes" or "contain", "containing", "contains" and are inclusive or open-ended terms that specifies the presence of what follows e.g. component and do not exclude or preclude the presence of additional, non-recited components, features, element, members, steps, known in the art or disclosed therein.

[0027] Furthermore, the terms first, second, third and the like in the description and in the claims, are used for distinguishing between similar elements and not necessarily for describing a sequential or chronological order, unless specified. It is to be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the invention described herein are capable of operation in other sequences than described or illustrated herein.

[0028] The recitation of numerical ranges by endpoints includes all numbers and fractions subsumed within that range, as well as the recited endpoints.

[0029] The expression "% by weight", "weight percent", "%wt" or "wt%", here and throughout the description unless otherwise defined, refers to the relative weight of the respective component based on the overall weight of the formulation.

[0030] Whereas the terms "one or more" or "at least one", such as one or more or at least one member(s) of a group of members, is clear per se, by means of further exemplification, the term encompasses inter alia a reference to any one of said members, or to any two or more of said members, such as, e.g., any >3, >4, >5, >6 or >7 etc. of said members, and up to all said members.

[0031] Unless otherwise defined, all terms used in disclosing the invention, including technical and scientific terms, have the meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. By means of further guidance, definitions for the terms used in the description are included to better appreciate the teaching of the present invention. The terms or definitions used herein are provided solely to aid in the understanding of the invention.Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Thus, appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment, but may. Furthermore, the particular features, structures or characteristics may be combined in any suitable manner, as would be apparent to a person skilled in the art from this disclosure, in one or more embodiments. Furthermore, while some embodiments described herein include some but not other features included in other embodiments, combinations of features of different embodiments are meant to be within the scope of the invention, and form different embodiments, as would be understood by those in the art. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0032] The term "image data" refers to digital information representing visual content in the form of pixels, typically organized in a two-dimensional matrix, where each pixel contains attributes such as color, intensity, or transparency, used to convey or analyze visual patterns, shapes, or objects. The image data can be captured by a plurality of different types of image sensors, such as, but not limited to 2D imagers, 3D imagers, infrared imagers, thermal imagers, hyperspectral imagers, line scan imagers, matrix cameras, ultrasound imagers, X-ray imagers.

[0033] The term "mycelium" refers to vegetative part of a fungus, consisting of a network of thread-like structures called hyphae, which grow through and absorb nutrients from the substrate, serving as the foundation for fungal growth and reproduction.

[0034] The term "hyphae" refers to thread-like, tubular structures that form the basic structural unit of a fungus, comprising a mycelium network and responsible for nutrient absorption, growth, and reproduction in fungal organisms.

[0035] The term "pin", "primordia" or "nodules" or "pin heads" refers to a small, rounded growth or formation within the mycelium, often serving as a precursor to mushroom fruiting bodies, formed as part of the fungal reproductive cycle under favorable environmental conditions. Typically, pins are defined as formations within the mycelium with a maximum diameter of about 5 mm, with formations of a higher diameter being classified as mushrooms.The term "modulation", "pin formation" or "budding" refers to the formation of pins or pinheads from the mycelium.

[0036] The term "reducing" or "removing" can refer to any action that stops the growth of a pin or mushroom, ranging from a specific action to kill the pin head / pin / mushroom (and hence stopping further growth), to partial or full destruction, and / or partial or full removal thereof from the mushroom bed.

[0037] The term "mushroom" refers to the fruiting body of a fungus, typically consisting of a stem (stipe), cap (pileus), and gills or pores, which produces and disperses spores as part of the fungal reproductive cycle.

[0038] The term "size representative parameter" refers to a physically detectable parameter that provides a direct or indirect indication of the objective size of an object, such as depth, width, length, height, diameter, circumference.

[0039] The term "environmental parameter" refers to parameters of the environment at or around the mushroom bed, which is typically inside. The environmental parameters may however also include outside conditions insofar as these impact the indoor parameters. Typical parameters include humidity, temperature (of soil and / or air), air flow, oxygen level, CO2 level, evaporation capacity, etc.

[0040] The term "humidity" refers to the amount of water vapor present in the air, typically expressed as a percentage of the maximum vapor capacity at a given temperature, which influences environmental conditions critical for biological and chemical processes, such as mushroom cultivation. The humidity can be described as an absolute value or as a relatively value. Additionally, this parameter can also be further characterized via the moisture deficit in the air, namely the amount of moisture that the air can still 'absorb' before saturation occurs.

[0041] In a first aspect, the invention relates to a computer-implemented method for monitoring mycelium, pin and mushroom growth (as well as the pin formation and outgrowth) in a mushroom bed via machine vision recognition, comprising the following steps:

[0042] a. acquiring image data of the mushroom bed;

[0043] b. individually detecting a plurality of pins and / or mushrooms in the image data and registering the position of each of said detected pins and / or mushrooms;c. determining one or more parameters associated to each of the detected pins and / or mushrooms from the image data, at least one of said parameters being representative of size of the pins and / or mushrooms;

[0044] d. tracking the one or more parameters for each of the individual pins and / or mushrooms over time;

[0045] e. proposing and / or initiating a reaction based on an analysis of said tracked parameters.

[0046] Up to this point in time, limited innovation using vision technology has taken place in the stages before picking. Vision technology has already been implemented in the picking stage up to a certain point, as it can be deployed in a fairly straightforward manner, namely a detection of a minimum size, shape, coloration, etc., or even a general 'level' of maturity for an entire bed, that would identify a "mature" mushroom that is ready for picking. It is in the growth stages however that the rules are not as well-defined, where the use of vision technology is not yet integrated.

[0047] The invention uses the abilities to individually detect mushrooms and / or pins in the image data to build up a database for each separate of these objects (mushrooms, pins and potentially mycelium), tracking a number of parameters for them, amongst which at least one size-representative parameter. While these parameters were historically used only in real-time for determining the time to pick the mushrooms, they can however be used in a more data-intensive fashion, looking at the evolution in some of these parameters to determine whether or not corrective or substantial actions are necessary (pin reduction / removal, adjustments to environmental parameters, ventilation, disease control, ...). None of the known methods employ historical data on individual level of pins and / or mushrooms to determine next steps, and realistically only look at a current value for a parameter to determine whether or not a (picking) action is to be undertaken.

[0048] A first focus lies on the detection of the individual objects and the determination of (individual) parameters of these individual objects, and using this parameters - after analysis - for further steps. The invention provides for an automated method that may either suggest further steps or actions, but can also take (certain) steps or actions autonomously. A combination of the two is also envisioned as part of the invention.

[0049] In a preferred embodiment, the analysis comprises individual growth tracking for each pin and / or mushroom that was individually detected, by registering the size-related parameter, or a derivative thereof, over time, and using this tracked individual growth as a parameter. Based on this, either an automated functionality can be provided that detects diverging growth patterns from the tracked individual growth and can alert operators and / or even take corrective measures, and / or it can simply be provided to an operator who can then determine themselves whether or not the growth diverges from a desired path.

[0050] In a preferred embodiment, the analysis comprises determining cluster parameters that are representative for multiple, spatially associated objects. Typically, these are located within predefined zones that may or may not be overlapping. The zones may be in a grid pattern, but can also have more organic shapes, such as circular. This allows the analysis to take into account both individual parameters as well as parameters that are representative for a subset of a population present in a certain zone. This allows a more general point of view that can present insights in growth in specific zones, but can also highlight issues based on location.

[0051] In a preferred embodiment, global parameters are determined associated to one or more bounded areas in the mushroom bed, said bounded areas comprising a cluster of multiple of the pins and / or mushrooms, wherein said global parameters are tracked overtime, with said global parameters representing a state of said bounded areas, and wherein the analysis further takes into account the global parameters. Such global parameters or cluster parameters can be used to determine locally averaged parameters and use these for further decision making.

[0052] Based on the cluster and global parameters, certain actions can be taken that are optimal cluster-wide or globally, as it is often difficult or even impossible to perform such actions on local scale. By aggregating the data, optimal further steps can be chosen that maximize yield / quality, for instance by maximizing the growth for pins / mushrooms in a certain size range to "invest" only in the worthwhile pins / mushrooms, thus optimizing the results.

[0053] In a preferred embodiment, the analysis comprises determining age group parameters that are representative for multiple, temporally associated objects. Typically, these are objects that entered a certain growth phase (for instance, nodulation, as this is usually the first point at which individual objects can be - more easily - detected) at a similar point in time, within a predefined time period. These time periods may be subsequent and separate, but may also be overlapping. The time periods may all be equal in length but can also vary, having multiple age groupspotentially nested in a larger age group. This allows the analysis to take into account both individual parameters as well as parameters that are representative for a subset of a population that share a certain time (and circumstance) of origin. Again, the general point of view can present new insights on growth but can also highlight issues at certain points in time, which can be taken into account for later (for instance, possibility that specific combinations of environmental parameters are less effective than previously estimated).

[0054] The global / cluster / age group parameters may be the same parameters as the tracked parameters, and can be an average of the individually tracked parameters, but may also be different parameters, for instance those only relevance over a larger population. Additionally, these global / cluster / age group parameters may further comprise statistical information on the set of individual parameters, such as variance and other statistical data.

[0055] Obviously, having cluster and age group parameters is a more preferred embodiment, leveraging the advantages of both of the above embodiments.

[0056] In a preferred embodiment, the tracked parameters comprise one or more of the following: mushroom / pin density per m2(preferably in a statistical representation, such as with P25-P50-P75), average size, average growth rate, average growth rate in association to size, largest size, largest size and associated growth relation, relative growth (with respect to size) over a predefined time period (for instance last 6h, 12h, 24h, etc.), comparison of an estimated growth with actual growth, population density over / under a predefined threshold of another (size-related) parameter, coarseness of mushroom (cap), color, shape.

[0057] In a preferred embodiment, the step of tracking the parameters starts (long) before the fruiting bodies of the mushrooms occurs, since at that point, the operator is limited in their actions, and can only time the picking operation sooner or later. The applicant has however found that most of the optimization can be found in the preceding stages, amongst which finetuning mycelium growth parameters that lead up to the nodulation stage, but also triggering, postponing, etc. of certain preceding stages. As such, the image data comprises images overtime of the mushroom beds, wherein location data is registered and associated to the image data, and positions therein, in order to link detected objects between separate image data.In preferred embodiments, earlier image data can be retroactively analyzed, based on the image data in which pins and / or mushrooms are detected, in order to determine optimal and less optimal conditions before these stages, namely during the mycelium growth stages. By individual detection of pins and / or mushrooms, these locations can be registered and used when analyzing the earlier image data, preferably along with environmental parameters at these earlier stages. Preferably, this earlier data relates to previous tides or harvest cycles. Optimally, the data can be clustered in such tides, which allows the system to more easily detect periodicity in the data. As such, the earlier image data can comprise at least the preceding tide, but preferably more than one preceding tides.

[0058] In a preferred embodiment, one or more events are registered with event data and logged with a time stamp, and associated to the mushroom bed and / or to the bounded areas within the mushroom bed, and / or to the detected pins and / or mushrooms within the mushroom bed, said events relating to a discontinuation of hydration and / or an airing or venting action (so-called "cooling down") and the associated start of nodulation / primordia formation, and wherein the analysis further takes the one or more events into account.

[0059] In the cultivation process, a number of events have a substantial impact, and can trigger the start of next phases, such as nodulation, etc. In order to be able to detect the positive or negative impact of timing of certain events on the cultivation process, a careful registration is necessary, in order to ascertain whether or not this is a correlation between the event and the result. Again, by tracking parameters over time, instead of single-shot measurements as in the prior art, it is possible to more reliably determine if a detected impact on the result is correlated to the event, thus allowing the system to learn for the feature whether or not the timing of the event was appropriate or not.

[0060] Most notably, the above events of stopping hydration and airing / venting action are crucial in triggering nodulation.

[0061] In a further preferred embodiment, the reaction is based on a subset of the parameters and optionally global parameters within a time period after the time stamp of the event data, said time period having a maximal duration of 3 days, and said time period being at least 12 hours after the time stamp of the event data. In order to detect the impact, image data needs to be removed over a time period up to 3 days from the event, as that is typically the longest until the nodulationshould be well under way or even complete. Preferably, sufficient image data would already be available after 48 hours or even 36 hours. However, a time period of at least 12 or even 24 hours is necessary to arrive at a well-founded analysis of the situation, and to definitively see patterns and tendencies in the nodulation and the growth of the pins / mushrooms. Only if sufficient image data is collected in which the pins and mushrooms can be detected and from which usable parameters can be determined and tracked (i.e., measured over a certain time period), conclusions can be drawn on whether or not the result was positive.

[0062] In a preferred embodiment, the step of individually detecting the plurality of pins and / or mushrooms is performed by a machine learning recognition model, wherein said recognition model is trained by supervised learning and / or reinforcement learning, wherein the recognition model is trained on a labeled mushroom dataset comprising a plurality of mushroom images in a mushroom bed, preferably under substantially similar conditions with respect to resolution, lighting and angle with respect to said mushroom bed as the image data, wherein mushrooms and / or pins in said mushroom images are labeled.

[0063] In an alternative embodiment, the recognition model is trained by unsupervised learning.

[0064] The recognition model is configured to generate a bounding region on the image data representative of the mushrooms and / or pins that are individually detected in said image data.

[0065] As discussed, one of the main issues in scaling up capacity of mushroom cultivation, is that the experience and expertise of the operators is difficult to transfer. Training is mainly on the job and the result of experience, rather than a formalized framework as it is difficult to objectively lay down frameworks. There is an enormous variety in cultivation between different strains, between different climates, ground types, and even in similar cultivation circumstances, many cultivators still differ in approach. One such example is the coverage of mycelium over the mushroom bed surface that is deemed sufficient to progress to the next phase of nodulation. General consensus is that optimal coverage is around 50%, but some cultivators wait until a coverage of more than 80% or even 90% is reached, while others go drastically below the 'ideal' 50%.

[0066] As a result, there is a strong need for objectively interpreting data. In order to arrive at this goal, machine learning is employed in the invention to generate a recognition model that automatically detects individual pins and / or mushrooms in the imagedata, in order to be able to derive individual parameters for each detected pin and / or mushroom, for tracking (the growth of) each mushroom / pin separately.

[0067] Ideally, this recognition model is trained by supervised learning and / or reinforcement learning.

[0068] Supervised learning offers several advantages when used to train a mushroom recognition model for identifying individual mushrooms or pins in a growth bed. By relying on labeled data, the model learns to accurately differentiate mushrooms or pins from other objects such as the substrate, debris, or contaminants. This ensures precise identification and allows the system to effectively detect individual mushrooms or pins, even in challenging scenarios where they are partially obscured or clustered closely together.

[0069] The approach also enhances model performance significantly. With enough labeled data, supervised learning algorithms, particularly convolutional neural networks (CNNs), can achieve high levels of accuracy, minimizing false positives and negatives. The ability to fine-tune the model with species-specific datasets further improves its reliability in diverse cultivation environments.

[0070] Supervised learning models are highly scalable and adaptable. They can be trained to recognize different mushroom species, making them suitable for a variety of cultivation setups. Additionally, they can be updated easily as new data becomes available, accommodating changes in cultivation practices or the introduction of new species.

[0071] In the context of mushroom cultivation, supervised learning facilitates automation by enabling real-time monitoring of growth beds. This reduces the need for manual labor and allows for precise growth analysis by measuring parameters such as size, density, and distribution of mushrooms or pins. Moreover, these models integrate well with loT systems and robotic applications, guiding automated harvesting, substrate assessment, or monitoring growth conditions. They can also contribute to yield prediction by combining visual data with environmental sensor inputs.

[0072] Another key advantage is the model's ability to detect anomalies, although this advantage is also applicable to unsupervised (or semi-supervised) learning techniques (for instance via trained autoencoders which are often used for anomaly detection). By training it to recognize healthy mushrooms or pins, the system can identify contamination, pest infestations, or other issues early in the cultivation process. It can also spot irregularities in growth, such as undersized or malformed mushrooms or pins, enabling timely intervention to optimize the yield.Reinforcement learning offers unique advantages when applied to training a recognition model for identifying individual mushrooms or pins in a growth bed. Unlike supervised learning, which relies on labeled datasets, reinforcement learning trains the model through interaction with its environment. This approach allows the system to learn optimal strategies for identifying mushrooms or pins by receiving feedback in the form of rewards or penalties based on its actions.

[0073] One of the key benefits is the model's ability to adapt to dynamic and complex environments. Mushroom growth beds often feature variations in lighting, substrate texture, and mushroom positioning. Reinforcement learning enables the model to continuously refine its decision-making process, ensuring robust recognition under diverse and changing conditions. This adaptability makes it particularly effective in real-world cultivation scenarios where conditions are not always uniform.

[0074] Another advantage lies in the potential for reinforcement learning to go beyond static identification tasks. The model can learn to prioritize certain actions, such as focusing on clusters of mushrooms or pins for improved efficiency or identifying mushrooms that are ready for harvesting. By optimizing its actions based on a cumulative reward system, the model can improve the overall performance of automated systems in tasks like selective harvesting or growth monitoring.

[0075] Reinforcement learning is also highly scalable and can integrate seamlessly with robotics and loT systems in mushroom cultivation. For example, a robot equipped with a reinforcement learning-based recognition model can adapt its movements and harvesting techniques to minimize damage to mushrooms or pins or the surrounding substrate. This level of adaptability enhances the precision and efficiency of automated cultivation processes.

[0076] Additionally, reinforcement learning supports long-term improvement. Unlike models that require periodic retraining with new labeled data, reinforcement learning systems can continue to learn and improve their performance over time through interaction with the environment. This makes them ideal for settings where ongoing optimization is needed to handle new challenges, such as detecting novel growth patterns, contamination, or unexpected changes in environmental conditions.

[0077] Lastly, reinforcement learning facilitates the development of intelligent systems capable of making context-aware decisions. By learning the relationships between environmental factors and mushroom or pins growth, the model can assist in optimizing cultivation strategies, such as adjusting lighting, humidity, or airflow, to enhance yields. This capability provides a comprehensive solution for automated mushroom cultivation, combining identification, monitoring, and optimization in a single framework.In a further preferred embodiment, the image data is processed for data augmentation with one or more of: blurring, scaling, binarization, random drop-out, noise reduction, cropping, contrast enhancement, color space manipulation.

[0078] Other augmentation techniques that can be used in addition or as an alternative, can include: resizing with or without aspect ratio maintenance, region of interest cropping, normalization, brightness normalization, gamma correction, artifact reduction, grayscale conversion, rotation alignment, perspective correction, sharpening, occlusion handling, channel rearrangement, batch normalization, upscaling, mean subtraction, variance scaling.

[0079] In order to train the recognition model, a limited set of training image data can be augmented to a more complete, larger dataset by performing one or more of the following steps on the available training images: blurring, scaling, binarization, random drop-out, noise reduction, cropping, contrast enhancement, color space manipulation, rotation to increase orientation invariance; translation to provide positional variance; flipping; shearing; brightness adjustment; color jittering; grayscale conversion; normalization; adding noise; sharpening; removing compression artifacts; masking to focus / remove focus; Mixllp; CutMix; grid masking; elastic distortion; random erasing.

[0080] In a preferred embodiment, said one or more parameters comprises an occupied area on at least a subarea of the mushroom bed. The subarea represents at least part of the surface of the mushroom bed, but can also extend over multiple beds, up to an entire growing cell (all of the beds in a space or even in an entire multi-room facility). The analysis comprises determining a total mushroom bed surface occupancy quantification for said subarea from said occupied area of the individual mushrooms and / or pins, with the total mushroom bed surface occupancy quantification representing the relative surface of the subarea of the mushroom bed being occupied by the detected mushrooms and / or modules. Finally, the analysis comprises tracking the change of said total mushroom bed surface occupancy quantification over time.

[0081] One of the key factors determining optimal growing conditions is the density with which the total surface is occupied. While this is correlated to the number of pins and mushrooms, the actual surface that is effectively occupied plays an important role in terms of aeration and exchange of moisture with the air. Furthermore, this parameter weighs heavily on the quantity and quality of the eventually harvested mushrooms. Areas with low densities lead to subpar yields in the same sense as too densely populated areas lead to bad yields. By keeping track of the surfaceoccupancy, the method can timely alert an operator and / or act autonomously when this exceeds predefined values. Typically, this mean that harvesting is necessary, pin removal is necessary, and / or environmental parameters need to be changed.

[0082] Preferably, the above mentioned occupancy quantification is performed only with regard to "living" mushrooms and / or pins (i.e., the ones that can still grow to a mature mushroom suitable for harvesting). The applicant has found that, building further on the preceding factor of the surface occupancy, the living surface occupancy, is even more crucial. This can be evaluated by tracking one or more of the individual parameters, in particular the size-related parameter, and determining if the parameters have changed according to an expectation or not in a predefined past period. Particularly, the speed of growth in a parameter of special interest. This period should of course be restricted in order to filter out mushrooms and pins that have just recently died off, while not being too strict and removing mushrooms and pins that are lagging or have temporarily slowed in their development.

[0083] In a preferred embodiment, wherein environmental parameters are periodically measured and associated to the image data, preferably wherein timestamps of image data and associated environmental parameters substantially align, wherein the analysis further takes into account said associated environmental parameters.

[0084] In a preferred embodiment, the environmental parameters comprise at least one, preferably two, more preferably three, and most preferably all, of: humidity, temperature, CO2 level and air flow at the mushroom bed. The temperature may be that of the air at the mushroom bed, but may also be that of the soil directly, such as of the soil directly at the top, the contact soil between the air and the mushroom or deeper-lying soil, or two or more of the above. Preferably, the temperature at the mushroom bed is that of the top layer of soil and / or that at the level of the pins / mushrooms themselves.

[0085] Additionally, external environmental parameters can also be registered, such as temperature, humidity, etc., outside of the direct environment, for instance outside. Usually, these have a small influence on the growth, as it tends to bleed through somehow, with the weight depending on the level of the cultivation operation. As such, the registering of these external environmental parameters may also assist in the analysis and detection of trends, patterns, and allow a more optimized cultivation to be achieved.In a preferred embodiment, the model provides for an optimized climate projection for the mushroom bed(s), based on real-time image data of said mushroom bed(s) and associated environmental parameters, and wherein said projection is based on the training data, along with environmental parameters associated to said training data. This can further take into account external environmental parameters as well as forecasts in such external environmental parameters (weather forecasts, for instance), which either can be inferred directly but may also be gathered from external sources directly, for instance via the internet.

[0086] In a preferred embodiment, the actions may include one or more of the following: adjustment of humidity, temperature, CO2 level and / or air flow.

[0087] In a preferred embodiment, the analysis comprises differentiating the individually detected pins and / or mushrooms into growing objects and non-growing objects, based on change over time of the at least one of said parameters being representative of size according to a predetermined threshold for said change, wherein information regarding said differentiated growing and / or non-growing objects is used for said reaction and / or presented to an authorized user.

[0088] In a further preferred embodiment, differentiating the individually detected pins and / or mushrooms into growing objects and non-growing objects is performed based on the changes over time in a period of at most 48 hours, preferably at most 24 hours, more preferably at most 12 hours, even more preferably at most 6 hours, even more preferably at most 3 hours and most preferably at most 1 hour, prior to the present.

[0089] In a preferred embodiment, the method comprises a step of tracking nodulation retroactively, by individually detecting the pins, and tracking said individually detected pins back to associated mycelium sections at the position of said individually detected pins in the image data of a time prior to the image data in which the pins are (first) individually detected, wherein said associated mycelium sections (and hyphae detected therein) are analyzed and parametrized in terms of flakiness, thickness, length, flakiness, size, number, stringiness, branching, color, and other features.

[0090] The above allows the method (and potentially associated model) to more accurately detect 'good' mycelium sections and 'bad', by being able to predict the pins that will develop therefrom. These can furthermore be annotated or labeled as not all pinsare positive (for instance, too small, too densely spaced, wrong color, disease, etc.), that can assist further in the finetuning of the method and model

[0091] In a further preferred embodiment, the method further comprises a step of tracking fruiting bodies of the mushrooms themselves as well retroactively, and tracing these back to pins and from there to mycelium sections at the position of the mushroom fruiting bodies in the image data of a time prior to the image data in which the fruiting bodies are individually detected, wherein said associated mycelium sections (and hyphae detected therein) are analyzed and parametrized in terms of flakiness, thickness, length, flakiness, size, number, stringiness, branching, color, and other features.

[0092] By tracing back the fruiting bodies to the mycelium, and potentially to intermediary pins, the method and associated model can be more accurately trained to detect 'good' and 'bad' mycelium in terms of the outcome in fruiting bodies. These can furthermore be annotated or labeled as not all fruiting bodies are positive (for instance, too small, too densely spaced, wrong color, disease, etc.), that can assist further in the finetuning of the method and model.

[0093] In a further aspect of the invention, the invention relates to a similar method as before, comprising the steps of:

[0094] a. acquiring image data of the mushroom bed;

[0095] b. (individually) detecting a plurality of hyphae of mycelium in the image data and registering the position of each of said detected hyphae, preferably via a demarcation in the form of a bounding box or region around said detected hyphae;

[0096] c. determining one or more parameters associated to the hyphae from the image data, preferably at least one of said parameters being representative of status of the hyphae, wherein the status defines whether or not the hyphae is growing per a predefined definition;

[0097] d. tracking the one or more parameters for the hyphae over time;

[0098] e. proposing and / or initiating a reaction based on an analysis of said tracked parameters.

[0099] As can be seen, in this variation, the method does not necessarily include a step of individually detecting a plurality of pins and / or mushrooms in at least some of the image data (although it may and preferably does, as the combination of multiple aspects improves the current inventions further). However, when focusing on the mycelium growth stage, there is no need to attempt to individually detect pins andmushrooms. In fact, prior to nodulation, the image data will mainly show soil and the growing mycelium. In these images, the mycelium is detected and localized in the image data, and can be individually detected in terms of hyphae threads, although this is not strictly necessary. More important than the individual parameters however, is the mycelium surface occupancy. This is less an individual parameter than a local parameter (cluster or global parameter) and will typically be defined for the totality of one or more images, spanning a zone that is usually part of a mushroom bed, but may also be an entire mushroom bed or multiple mushroom beds.

[0100] Parameters that can be tracked individually for the hyphae can be thickness, length and other size-related parameters, as well as flakiness (size, number, etc.), stringiness, branching, color, and others. Again, these can then be tracked as age group / global / cluster parameters, for further use in analysis.

[0101] In a preferred embodiment, the image data is acquired along a line of sight substantially perpendicular to a plane defined by the mushroom bed, and the individual detection of pins is executed with a first machine learning model, and wherein the individual detection of mushrooms is performed with a second machine learning model different from the first machine learning model. The substantially perpendicular imaging reduces perspective distortion and enables a more accurate determination of size-related parameters, which are essential for reliable tracking and analysis. Furthermore, by employing separate machine learning models for pins and mushrooms, the detection process can be specialized for objects at different growth stages, thereby improving detection accuracy, particularly for small and low-contrast pins that are otherwise difficult to detect.

[0102] In alternative embodiments, the image data may be acquired at an angle deviating from the perpendicular orientation within a predefined tolerance range, for instance within ±10° to ±20°, while still maintaining sufficient accuracy. Additionally or alternatively, a single machine learning model may be used, optionally configured with stage-dependent parameters or classification thresholds. It is further envisaged that the first and second models may differ in architecture, training data, or parameterization rather than being fundamentally different in type.

[0103] Preferably, both models operate on the same image data, although in some embodiments different preprocessing steps may be applied prior to detection.In a preferred embodiment, the individually detected pins or mushrooms are separated based on diameter. Objects with a diameter higher than 5 mm are classified as mushrooms, below as nodules (and if no individual objects are identified, this is typically classified as mycelium). This introduces an objective and reproducible classification criterion, enabling automated differentiation between growth stages without reliance on subjective interpretation. It furthermore improves consistency and allows downstream processes, such as harvesting decisions or pin removal, to be executed more reliably.

[0104] In alternative embodiments, the threshold may be defined within a range, for instance between 3 mm and 10 mm, or may be dynamically adjusted based on mushroom species or environmental parameters. The classification may further take into account additional parameters, such as shape, growth rate, or texture, in combination with the diameter.

[0105] Preferably, the diameter is determined based on image-derived parameters, such as an equivalent circular diameter or a bounding region, optionally calibrated to physical units.

[0106] In a preferred embodiment, the first and / or second machine learning model comprise a U-Net type model. This is particularly advantageous in the context of mushroom cultivation, where objects may be densely clustered and exhibit complex biological textures.

[0107] In alternative embodiments, other segmentation architectures may be employed, such as encoder-decoder networks or other pixel-wise classification models. The term "U-Net type model" is intended to encompass architectures comprising contracting and expanding paths with feature fusion, such as skip connections.

[0108] In a further preferred embodiment, the first and / or second machine learning model comprise a U-Net type model and a subsequent Mask Region-based Convolutional Neural Network (R-CNN) model, wherein the U-Net type model performs a pixel-by-pixel evaluation of the image data to determine a likelihood of a pin- or mushroom pixel, and wherein the Mask R-CNN model subsequently performs an individual pin or mushroom instance detection on the image data augmented with the pixel-by-pixel evaluation. This combination integrates semantic segmentation with instance detection, thereby improving the ability to distinguish overlapping or closely spaced objects and reducing false detections. In particular for colored types of mushrooms(chestnut mushrooms and the likes), it has been found that this combination performs exceptionally well.

[0109] In alternative embodiments, other two-stage pipelines may be used, wherein a first model generates a pixel-wise or region-based representation and a second model performs object-level detection. The augmentation of the image data may comprise adding an additional channel or feature map representing the pixel-wise likelihood.

[0110] In a preferred embodiment, an individual growth metric is determined associated to each pin and / or mushroom, wherein a growth map is generated based on the image data and wherein the pins and / or mushrooms are visually marked based on the individual growth metric, wherein a first type of visual marker is applied to pins and / or mushrooms of which the individual growth metric is above a predefined first threshold value, and wherein a second type of visual marker is applied to pins and / or mushrooms of which the individual growth metric is below a predefined second threshold value, the second threshold value being equal to or below the first threshold value, and wherein the individual growth metric is based on a time-based evolution of the parameter representing the size. This enables intuitive visualization of growth dynamics, allowing rapid identification of underperforming or overperforming objects or zones, and facilitating timely intervention.

[0111] In alternative embodiments, more than two thresholds may be used, or the output may be provided in a non-visual form, such as numerical rankings or alerts. The growth metric is preferably based on a time-dependent evolution of a size-related parameter, such as a relative growth rate over a predefined time window.

[0112] The visual marker may be via annotations, coloring, etc., added to images of the mushroom beds, or can even be directly applied to the objects themselves, for instance by lights (LED, laser, etc.) to indicate objects that fall below certain thresholds.

[0113] In a preferred embodiment, the method further comprises a step of predicting a suitable harvesting time for one or more of the mushrooms based on the tracked parameters. This enables forward-looking optimization of the cultivation process, improving yield quality and allowing more efficient planning of harvesting operations.In alternative embodiments, the method may predict a harvesting window or a maturity level rather than a specific time. The prediction may be performed at the level of individual mushrooms, clusters, or entire beds.

[0114] Preferably, the prediction is based on historical and real-time growth data, optionally in combination with environmental parameters.

[0115] In a preferred embodiment, the method further comprises a step of alerting when a suitable harvesting time is applicable for one or more of the mushrooms based on the tracked parameters. This enables timely intervention without requiring continuous manual monitoring, thereby reducing reliance on operator expertise and enabling integration with automated systems.

[0116] In alternative embodiments, the alert may be generated based on predefined thresholds or prediction confidence levels, and may be communicated via various interfaces or directly trigger control actions in the cultivation system.

[0117] Preferably, the alert is directly linked to the predicted harvesting time derived from the tracked parameters.

[0118] In a preferred embodiment, the image data comprises digitally encoded image files, preferably compressed image files, and wherein the image data is combined with sensor data by means of sensor fusion, said sensor data comprising one or more of: air temperature, compost temperature, air humidity and CO2 level, and wherein the analysis is performed based on the fused image data and sensor data. This allows the system to correlate visual growth information with environmental conditions, thereby improving the robustness and accuracy of the analysis and predictions.

[0119] In alternative embodiments, only a subset of environmental parameters may be used, or the fusion may be performed at different stages, such as at feature level or decision level.

[0120] Preferably, the sensor data is associated with the image data based on temporal correspondence.

[0121] In a further preferred embodiment, the sensor fusion comprises associating the sensor data with the image data based on corresponding timestamps and / or spatial alignment, and integrating the sensor data as additional input features for theanalysis and / or for one or more of the machine learning models. This ensures correct correlation between environmental conditions and observed growth, thereby preventing misinterpretation due to temporal or spatial mismatch.

[0122] In alternative embodiments, approximate alignment within a predefined tolerance may be sufficient. The integration of the sensor data may be performed by incorporating it as numerical input features or as additional channels in the model input.

[0123] Preferably, the alignment ensures that the environmental parameters correspond to the same physical region and time instance as the image data.

[0124] In a preferred embodiment, at least one of the parameters that is tracked is a status-related parameter that allows the hyphae to be evaluated on whether it is considered living (growing) or not. This status-related parameter can be size-related (such as thickness), where the growth of this parameter over time is monitored, at least over a predefined last time period, to see if the hyphae is still growing. Alternatively, the status-related parameter can focus on the tendency to branch, or other parameters that allow it to be defined as living or not.

[0125] In a preferred embodiment, the one or more parameters for the mycelium comprise an occupied area on at least a subarea of the mushroom bed. The subarea represents at least part of the surface of the mushroom bed. The analysis comprises determining a total mycelium surface occupancy quantification for said subarea from said occupied area of the mycelium, with the total mycelium surface occupancy quantification representing the relative surface of the subarea of the mushroom bed being occupied by the detected mycelium. Finally, the analysis comprises tracking the change of said total mycelium surface occupancy quantification over time.

[0126] One of the key factors determining optimal growing conditions is the density with which the total surface is occupied by mycelium. While this is correlated to the number of hyphae, the actual surface that is effectively occupied plays an important role in terms of aeration and exchange of moisture with the air. By keeping track of the mycelium surface occupancy, the method can timely alert an operator and / or act autonomously when this exceeds predefined values. Typically, this mean that environmental parameters need to be changed, that (discontinuation of) hydration is required, that spraying is required to counteract (overly) rapid mycelium growth or other actions.Preferably, the above mentioned mycelium occupancy quantification is performed only with regard to "living" mycelium (i.e., growing mycelium). The applicant has found that, building further on the preceding factor of the mycelium surface occupancy, the living mycelium surface occupancy, is even more crucial. This can be evaluated by tracking one or more of the individual parameters, in particular the size-related parameter, and determining if the parameters have changed according to an expectation or not in a predefined past period. This period should of course be restricted in order to filter mycelium that have just recently died off, while not being too strict and removing mycelium hyphae that are lagging or have temporarily slowed in their development. Using this information, it is a further objective of the invention to predict or forecast the living mycelium surface occupancy, and if necessary, take actions to counter unwanted variations in these predicted values to optimize the effective values.

[0127] In a preferred embodiment, the analysis comprises determining cluster parameters that are representative for multiple, spatially associated hyphae. Typically, these are located within predefined zones that may or may not be overlapping. The zones may be in a grid pattern, but can also have more organic shapes, such as circular. This allows the analysis to take into account both individual parameters as well as parameters that are representative for a subset of a population present in a certain zone. This allows a more general point of view that can present insights in growth in specific zones, but can also highlight issues based on location.

[0128] In a preferred embodiment, global parameters are determined associated to one or more bounded areas in the mushroom bed (which bounded area may include the entire bed), said bounded areas comprising a cluster of multiple of the hyphae, wherein said global parameters are tracked over time, with said global parameters representing a state of said bounded areas, and wherein the analysis further takes into account the global parameters.

[0129] Such global parameters or cluster parameters can be used to determine locally averaged parameters and use these for further decision making.

[0130] In a preferred embodiment, the analysis comprises determining age group parameters that are representative for multiple, temporally associated hyphae. Typically, these are hyphae that were first detected at a similar point in time, within a predefined time period. These time periods may be subsequent and separate, but may also be overlapping. The time periods may all be equal in length but can alsovary, having multiple age groups potentially nested in a larger age group. This allows the analysis to take into account both individual parameters as well as parameters that are representative for a subset of a population that share a certain time (and circumstance) of origin. Again, the general point of view can present new insights on growth but can also highlight issues at certain points in time, which can be taken into account for later (for instance, possibility that specific combinations of environmental parameters are less effective than previously estimated).

[0131] The global / cluster / age group parameters may be the same parameters as the tracked parameters, and can be an average of the individually tracked parameters, but may also be different parameters, for instance those only relevance over a larger population. Additionally, these global / cluster / age group parameters may further comprise statistical information on the set of individual parameters, such as variance and other statistical data.

[0132] Obviously, having cluster and age group parameters is a more preferred embodiment, leveraging the advantages of both of the above embodiments.

[0133] In a preferred embodiment, the tracked parameters comprise one or more of the following: mycelium density per m2(preferably in a statistical representation, such as with P25-P50-P75), average size-related parameters, average growth rate, average growth rate in association to size, relative growth (with respect to size) over a predefined time period (for instance last lh, 3h, 6h, 12h, 24h, etc.), comparison of an estimated growth with actual growth, color, shape.

[0134] In a preferred embodiment, one or more events are registered with event data and logged with a time stamp, and associated to the mushroom bed and / or to the bounded areas within the mushroom bed, and / or to the detected hyphae within the mushroom bed, said events relating to a discontinuation of hydration and / or an applied (forced) change in environmental parameters and / or airing / venting action and the associated start of nodulation, and wherein the analysis further takes the one or more events into account.

[0135] In the cultivation process, a number of events have a substantial impact, and can trigger the start of next phases, such as nodulation, etc. In order to be able to detect the positive or negative impact of timing of certain events on the cultivation process, a careful registration is necessary, in order to ascertain whether or not this is a correlation between the event and the result. Again, by tracking parameters overtime, instead of single-shot measurements as in the prior art, it is possible to more reliably determine if a detected impact on the result is correlated to the event, thus allowing the system to learn for the feature whether or not the timing of the event was appropriate or not.

[0136] Most notably, the above events of stopping hydration and airing / venting action are crucial in triggering nodulation.

[0137] In a further preferred embodiment, the reaction is based on a subset of the parameters and optionally global parameters within a time period after the time stamp of the event data, said time period having a maximal duration of 3 days, and said time period being at least 12 hours after the time stamp of the event data. In order to detect the impact, image data needs to be removed over a time period up to 3 days from the event, as that is typically the longest until the nodulation should be well under way or even complete. Preferably, sufficient image data would already be available after 48 hours or even 36 hours. However, a time period of at least 12 or even 24 hours is necessary to arrive at a well-founded analysis of the situation, and to definitively see patterns and tendencies in the nodulation and the growth of the pins / mushrooms that grow from the detected hyphae. Only if sufficient image data is collected in which the pins and mushrooms can be detected and from which usable parameters can be determined and tracked (i.e., measured over a certain time period), conclusions can be drawn on whether or not the result was positive.

[0138] In this embodiment, a combination is synergistically made with the method according to the first aspect, wherein the mushrooms and / or pins were individually detected and tracked in terms of parameters.

[0139] In a preferred embodiment, the method comprises a step of differentiating mycelium from non-mycelium in the image data of the mushroom bed (i.e., detecting the hyphae), and determining a coverage ratio of the mycelium for at least a subarea of the mushroom bed, wherein said subarea represents at least part of the surface of the mushroom bed, the coverage ratio representing the relative surface of the subarea being occupied by the differentiated mycelium in the subarea, and wherein the analysis comprises tracking the change of the coverage ratio over time.

[0140] In a preferred embodiment, said step of differentiating mycelium is performed based on color, stringiness and / or flakiness of the image data. Alternatively or additionally, the differentiating can be based on the change in other tracked parameters, such assize-related parameters. Further parameters that can be used to differentiate mycelium can be indirect, such as temperature, air humidity, etc. Positions without mycelium typically have higher evaporation of water, resulting in a different temperature than positions where mycelium is growing.

[0141] In a further preferred embodiment, the method further comprises a step of predicting a future mycelium density or population in the subarea based on the differentiated mycelium in the subarea, wherein said step of predicting is performed by a machine learning mycelium growth prediction model, wherein said mycelium growth prediction model is trained by supervised learning and / or reinforcement learning, wherein the mycelium growth prediction model is trained on a mycelium dataset comprising multiple sets of time sequence mycelium images, wherein each set comprises temporally sequential images of an at least partly overlapping subarea in a mushroom bed during mycelium growth, wherein the mycelium in the temporally sequential images is differentiated with respect to non-mycelium.

[0142] In an even further preferred embodiment, the mycelium growth prediction model is configured for, based on the image data for a subarea of the mushroom bed, generating a predicted future mycelium density image and / or predicted future mycelium density characteristics for said subarea, preferably wherein said image data for the subarea is processed for quality augmentation with one or more of: blurring, scaling, binarization, random drop-out, noise reduction, cropping, contrast enhancement, color space manipulation.

[0143] In a preferred embodiment, environmental parameters are periodically measured and associated to the image data, preferably wherein timestamps of image data and associated environmental parameters substantially align, wherein the analysis further takes into account said associated environmental parameters.

[0144] In a preferred embodiment, environmental parameters are associated to the mycelium images of the mycelium dataset, wherein timestamps of the mycelium images and the associated environmental parameters substantially align, and wherein the mycelium growth prediction model is trained taking into account said associated environmental parameters, wherein the step of predicting the future mycelium density or population further takes into account the environmental parameters associated to the image data for which the mycelium growth prediction model predicts the future mycelium density or population.In a preferred embodiment, the environmental parameters comprise at least one, preferably two, more preferably three, and most preferably all, of: humidity, temperature, CO2 level and air flow at the mushroom bed. The temperature may be that of the air at the mushroom bed, but may also be that of the soil directly, or both. Particularly, the temperature at the mushroom bed is that of the top layer of soil, or that at the level of the pins / mushrooms themselves.

[0145] Additionally, external environmental parameters can also be registered, such as temperature, humidity, etc., outside of the direct environment, for instance outside. Usually, these have a small influence on the growth, as it tends to bleed through somehow, with the weight depending on the level of the cultivation operation. As such, the registering of these external environmental parameters may also assist in the analysis and detection of trends, patterns, and allow a more optimized cultivation to be achieved.

[0146] In a preferred embodiment, the analysis comprises differentiating the individually detected hyphae into growing objects and non-growing objects, based on change over time of the at least one of said parameters being representative of size according to a predetermined threshold for said change, wherein information regarding said differentiated growing and / or non-growing objects is used for said reaction and / or presented to an authorized user.

[0147] In a further preferred embodiment, differentiating the individually detected pins and / or mushrooms into growing objects and non-growing objects is performed based on the changes over time in a period of at most 48 hours, preferably at most 24 hours, more preferably at most 12 hours, even more preferably at most 6 hours, or even at most 3, 2 or 1 hour prior to the present.

[0148] In a preferred embodiment, the model provides for an optimized climate projection for the mushroom bed(s), based on real-time image data of said mushroom bed(s) and associated environmental parameters, and wherein said projection is based on the training data, along with environmental parameters associated to said training data. This can further take into account external environmental parameters as well as forecasts in such external environmental parameters (weather forecasts, for instance), which either can be inferred directly but may also be gathered from external sources directly, for instance via the internet.

[0149] In a preferred embodiment, the actions may include one or more of the following: adjustment of humidity, temperature, CO2 level and / or air flow.In a preferred embodiment, the method and potentially associated model may be configured for determining an optimal mycelium occupancy quantification before triggering a next phase, namely nodulation, or proposing the next phase to be triggered. Again, determining when this happens is based on historical image data with which the model and method are trained, and allows interpretation of the current, typically real-time image data. Most preferably, the historical image data and / or current image data is further provided with accompanying environmental parameters (most preferably also external).

[0150] The model can, based on historical image data, compare a current situation based on (real-time) image data of the mushroom beds, with both image data sets preferably being augmented by further environmental parameters, and potentially event logs. By this comparison, the model can determine which historical situations are relevant and applicable to the current situation, by determining degrees of similarity between the real-time and historical image data, and determine optimal strategies based on the results with those historical situations.

[0151] For instance, a number of similar scenarios may be detected which exceed a similarity threshold to the current situation, and the model may then determine the 'success' rate in each of those scenarios (for instance, defined by a good yield at harvest), and base its proposed or initiated actions or reactions on the reactions and actions that were taken in those scenarios, in order to maximize the current success rate. The model can base its proposed or initiated (re)actions based on a combination of high similarity and high success rate, in an attempt to find a "sweet spot" historical scenario, and provide a (re)action that approaches the (re)action used in that sweet pot historical scenario, or it may use a hybrid approach and take a combination of multiple scenarios into account with high similarity and high success rate and provide a mixed (re)action based on the (re)actions in those scenarios. This mixed approach may be an average of the (re)actions of the multiple scenarios, but may also be a combination of (re)actions of multiple scenarios.

[0152] Typically, nodulation or pin formation is triggered based on a user-specific threshold. Some cultivators start nodulation at 50% coverage with mycelium, but as mentioned, others start at a much lower threshold, while others are known to go up to a 80 or 90% coverage. While such preferences may work in some occasions because of the high level of niche expertise in those extremes (the cultivator knowing exactly how to work in those strongly divergent circumstances that allow it to be successful), or even the available time for the cultivator who might not be able to check each mushroom bed regularly, there is a strong need for a more generalizedmethod of operation that does not require extreme expertise or constant supervision of an expert. Instead, an optimized strategy can be determined that can be implemented on a large scale, with a relatively low need for supervision or expertise. Usually, the mushroom beds are checked at least 2 times per day, but preferably more than that, such as 5 times per day. If the cultivator does not check two times per day for instance, the mycelium can experience such rapid growth in the time between checks that the actual coverage exceeds the desired coverage strongly, resulting in a subpar yield.

[0153] However, via the analysis and triggering based on user-specific thresholds for the coverage of the mushroom bed, the cultivator can substantially save time and reduce the need for checks to only once each day or even less, as the system will alert them timely.

[0154] The similarity threshold may be determined by a comparison of as many factors as possible, where each factor has an associated weight to determine a weighted score that is part of a similarity score for a historical image data set that represents a historical situation. For instance, external (outside) temperature will have a relatively low weight as opposed to air temperature at the mushroom bed (inside the cultivation room or growing room), so a strong similarity in external temperature will have a limited effect on the overall similarity score.

[0155] In a preferred embodiment of any of the above aspects of the invention, the reaction may comprise one or more changes that are effected in climate control, such as changing relative humidity, changing hydration of the mushroom bed, changing air temperature, changing soil temperature, changing air flow speed, changing air pressure, changing CO2 level, changing settings for determining whether or not a reaction is to be proposed and / or initiated.

[0156] Based on historical data, the method and associated model are trained to recognize desired patterns and evolutions in the tracked parameters, that allow them to initiate or propose actions at opportune moments, either to counteract a negative pattern / evolution (for instance, countering a too rapid growth of mycelium), or to build further on a positive pattern / evolution (detecting suitable conditions to trigger a next phase).

[0157] It is provided that the method and model can propose and / or initiate these reactions autonomously due to training based on the historical data, and apply this on realtime image data.In one such embodiment, it is envisioned that the settings can be changed for determining the triggering of a reaction proposal / initiation. This can for instance be that normally an action is triggered by a certain threshold being reached, but that this threshold can be modified in light of the current image data, such as detection of very fast growth of mycelium that cannot be curtailed by extra hydration (especially as this usually leads to a reduction in yield and quality) as would usually be done, but instead results in an accelerated airing / venting action to start nodulation.

[0158] In a preferred embodiment of any of the above aspects of the invention, the image data is gathered by one or more stationary image sensors provided at the mushroom beds. Preferably, these image sensors cover the entire mushroom bed. More preferably, in case of multiple image sensors, the field of view thereof is partly overlapping, such that certain sections of the mushroom beds are present in multiple fields of view.

[0159] Alternatively, the image data is gathered by movable image sensors that move over the mushroom beds in order to gather image data thereof. Preferably, this movement is periodical to gather image data for each position regularly.

[0160] The movable image sensors can be mounted via rails, allowing a very simple guidance of the image sensors along a fixed trajectory, while also allowing an easy way to powerthe image sensors and to gatherthe image data via a wired connection, although wireless is of course also possible.

[0161] Alternatively, the movable image sensors can also take other forms, such as via movable robot arms, via drones, etc.

[0162] In a preferred embodiment of any of the above aspects of the invention, the image sensors are mounted above the mushroom beds and have a substantially vertically downward orientation, as this provided the best and most objective view of the size of the cap, and can easily be processed to correct for object size discrepancies. Alternatively or additionally, the image sensors can be mounted in a skewed orientation.

[0163] In a preferred embodiment of any of the above aspects of the invention, the image sensors are one or more of 2D imagers, 3D imagers, infrared imagers, thermal imagers, hyperspectral imagers, line scan imagers, matrix cameras, ultrasound imagers, X-ray imagers.In a preferred embodiment on any of the above two aspects, the invention relates to an improvement in climate control, using the historical image data and associated environmental parameters, to determine an optimized climate control, wherein the analysis further comprises determining a proposed or initiated reaction in which climate control parameters (environmental parameters) are changed at the mushroom bed(s), based on the analysis, and wherein said changed climate control parameters relate to one or more of (relative) humidity, temperature of soil or compost (top layer) in the mushroom bed and / or of air at the mushroom bed, CO2 level and / or air flow. Temperature of the soil / compost is typically regulated indirectly via adapting the air temperature and / or air flow.

[0164] In a further aspect, the invention relates to a method for reducing or removing pins and / or mushrooms, preferably only pins, in a mushroom bed. The goal of this is to reduce the number of pins or nodules growing into mature mushrooms, to avoid competition for resources (space, moisture, nutrients, etc.).

[0165] The method comprises the steps of:

[0166] a. obtaining parameters of the pins and / or mushrooms in the mushroom bed, based on image data and optionally environmental parameters for the mushroom bed, wherein the parameters comprise one or more of:

[0167] a. spacing between the detected pins and / or mushrooms,

[0168] b. growth rate of the detected pins and / or mushrooms,

[0169] c. shape of the detected pins and / or mushrooms,

[0170] d. color of the detected pins and / or mushrooms, and

[0171] e. size differences between the detected pins and / or mushrooms; b. analyzing the data using a trained artificial neural network;

[0172] c. calculating a selection criterion for removing specific mushrooms based on one or more of:

[0173] a. potential to increase yield per area unit, and

[0174] b. an auto-encoder method that predicts the growth success rate of mushrooms based on the real-time data;

[0175] d. generating output in the form of instructions for reducing or removing specific detected pins and / or mushrooms in the mushroom bed.

[0176] As mentioned before, the neural network is trained based on historical image data and optionally associated environmental parameters, that allows the network to predict viability of the current situation, and based on this, propose pin removal steps.The determination of which pins (and / or modules) are to be removed is preferably based on more than one of the mentioned parameters. Typically, clustering is also taken into account, if for instance N pins are grouped together in a predetermined surface A, 1 or more will be removed. If more than 1 pins are removed, these are typically spread apart over the cluster. Note that only the number of "living" (i.e., growing) pins is counted to determine if removal / reduction is necessary.

[0177] In terms of spacing, ideally a sweet spot is found in a dense enough clustering that ensures a large yield, but not too dense that the resource competition would result in substandard fruiting bodies, allowing an optimal pin removal in terms of crowding. In terms of growth rate, this strongly determines the quality of the future fruiting body in size. By being able to remove pins with substandard growth, particularly in combination with the above density of clustering, the ideal pin removal 'victims' are determined. In terms of shape and color, this also is a strong predictor towards quality, again optimizing the pin removal selection. Finally, the size differences may be a result of a difference in nodulation timing. Combining this for instance with knowledge of the growth rate, allows further optimization of the pin removal selection.

[0178] In a preferred embodiment, the artificial neural network is trained based on historical datasets obtained from prior cultivation cycles. As mentioned, the historical datasets typically comprise image data and may or may not comprise additional associated environmental parameters, events, etc.

[0179] In a preferred embodiment, the selection criterion prioritizes the removal of pins (and / or mushrooms) with irregular shapes, non-uniform colors, or growth rates deviating from a predefined optimum.

[0180] In a preferred embodiment, the auto-encoder method is utilized to identify clusters of pins exhibiting a heightened risk of growth competition.

[0181] In a preferred embodiment, the method further comprises a step of updating the artificial neural network with new data collected during the pin removal process, enabling continuous improvement of pin removal accuracy.

[0182] In a further aspect, the invention relates to a system for removing pins and / or mushrooms, preferably pins, from a mushroom bed, preferably based on instructions from a method for pin removal as discussed in this document. The system comprises a pin removal tool.Pin reduction or pin removal can be the complete or partial removal of the pin and / or mushroom, but may also be the killing of the pin and / or mushroom, in terms of stopping further growth, and preferably even causing reduction of the size (for instance, due to moisture leakage or evaporation).

[0183] The pin removal tool may be movable via a suspension system over the mushroom beds, such as a rail system, which can allow movement in 2D or 3D. This can be performed via separate actuators, first moving the tool in a horizontal plane, with a second moving in the vertical direction (lowering and lifting the tool).

[0184] It may alternatively be movable via a robotic arm. The pin removal tool may be movable via translation and / or rotation, in order to allow the pin removal tool to cover a substantially large surface. This may be the entire mushroom bed or even multiple mushroom beds, or a part of a mushroom bed.

[0185] In some embodiments, the pin removal tool may be stationary, with its area of effect on the mushroom bed being movable by rotation or other manipulations (such as removal / insertion of lens, blockage of pathways, etc.). Again, this would allow the pin removal tool to cover a substantially large surface. This may be the entire mushroom bed or even multiple mushroom beds, or a part of a mushroom bed.

[0186] In a preferred embodiment, the pin removal tool comprises one or more vacuum tools with an opening that is moved to a position of a pin or mushroom to be reduced or removed and brought into close proximity of said pin or mushroom. The vacuum tool is in communication with a vacuum exerting means that provides a negative pressure at the opening of the vacuum tool, in order to pull the pin or mushroom out of the mushroom bed. It can then be removed either via a removal channel, through or via the vacuum tool, or it can be held at the opening and removed by moving the vacuum tool away.

[0187] The vacuum tool may have more than one openings at its end, in order to provide for a spread out deployment of the negative pressure. This allows a sufficient force to be employed, but less concentrated. This prevents the tool from destroying the pin or mushroom and causing debris (or even damaging / pulling in neighboring objects or soil), as opposed to removing the object to be reduced from the mushroom bed and removing it.

[0188] Preferably, the vacuum tool would be moved via translation via a robotic arm or via a rail and lowered to the desired position. Preferably, the angle of approach is directlyvertically downward, although other angles of approach are possible, or even a combination of multiple angles, via multiple vacuum tools.

[0189] In a preferred embodiment, the pin removal tool comprises one or more laser tools, configured for killing the pin and / or mushroom via a laser beam.

[0190] Preferably, the laser tool would be moved via rotation, by rotating the laser tool and direct the laser beam it generates at the targeted pin or mushroom.

[0191] Additionally or alternatively, the laser tool may also be moved via a translation (as said, potentially along with the rotation to cover an even large surface on which it can act, as well as providing more options for angle of attack on the object to be pruned). This can be performed via translation via a robotic arm or via a rail.

[0192] When moving only with a translation, the angle of attack is directly vertically downward, although other angles of attack are possible, or even a combination of multiple angles, via multiple laser tools.

[0193] If the laser tool is moved via rotation, the angle of attack can vary, although a threshold angle is to be defined to avoid damage to other pins / mushrooms, as well as the mushroom bed. Such an angle would preferably be at most 70°, even more preferably at most 60°, even more preferably at most 50°, even more preferably at most 40°, with respect to the vertical axis.

[0194] In a preferred embodiment, the pin removal tool comprises one or more vibrating tools configured to vibrate at high frequency laterally and / or longitudinally. The vibrating tool is in communication with a power source.

[0195] The vibrations are preferably limited in amplitude to the dimensions of the object to be pruned, or in general an average size thereof, preferably with a predefined corrective factor (for instance x 1.5, x 2, x 3, etc.) to ensure destruction without damaging neighboring objects.

[0196] Preferably, the vibrating tool would be moved via translation via a robotic arm or via a rail and lowered to the desired position. Preferably, the angle of approach is directly vertically downward, although other angles of approach are possible, or even a combination of multiple angles, via multiple vibrating tools.

[0197] In a preferred embodiment, the pin removal tool comprises one or more heating tools configured to locally generate heat at at least part of its surface, in order to kill the targeted pin or mushroom. The heating tool is in communication with a power source.The heating tool may be configured for contacting the surface of the targeted pin or mushroom and act upon the contacted surface, or may be inserted in the targeted pin or mushroom to act internally.

[0198] Preferably, the heating tool would be moved via translation via a robotic arm or via a rail and lowered to the desired position. Preferably, the angle of approach is directly vertically downward, although other angles of approach are possible, or even a combination of multiple angles, via multiple heating tools.

[0199] In a preferred embodiment, the pin removal tool comprises one or more air flow generation tools or air insertion tools configured to generate a high pressure, preferably unidirectional air flow or air stream, in order to kill the targeted pin or mushroom. Alternatively, the airflow is not unidirectional. The air flow / insertion tool is in communication with a power source.

[0200] The tool may be configured for approaching or even contacting the surface of the targeted pin or mushroom and act upon the contacted surface, destroying the pin or mushroom via a high pressure air flow that is generated aimed at the pin or mushroom.

[0201] Preferably, the tool may be inserted in the targeted pin or mushroom to act internally, and essentially inflate the pin or mushroom from the inside, again destroying it. In such a case, the generated air flow need not be unidirectional, and the tool can have a plurality of openings through which multiple separate air flows are provided.

[0202] The tool preferably has a head with one or more openings through which the air flow can be generated and is in communication with a high pressure air flow generation means capable of generating a high pressure air flow to the tool and through the one or more openings.

[0203] Preferably, the air flow generation or insertion tool would be moved via translation via a robotic arm or via a rail and lowered to the desired position. Preferably, the angle of approach is directly vertically downward, although other angles of approach are possible, or even a combination of multiple angles, via multiple tools.

[0204] In a preferred embodiment, the pin removal tool comprises one or more sharp tools configured to stab, tear, or otherwise damage the targeted pin or mushroom bylocalized pressure exertion on the outer surface (and subsequently inner volume), in order to kill the targeted pin or mushroom. The sharp tool may comprise one or more of a blade, knife, syringe, pointy instrument, barbed instrument, saw. These may or may not be motorized, for instance, to rotate.

[0205] Preferably, the tool is configured to destroy the pin or mushroom by repetitive movements up and down and / or laterally and / or rotationally.

[0206] Preferably, the sharp tool would be moved via translation via a robotic arm or via a rail and lowered to the desired position. Preferably, the angle of approach is directly vertically downward, although other angles of approach are possible, or even a combination of multiple angles, via multiple sharp tools.

[0207] In a preferred embodiment, the pin removal tool comprises one or more blunt tools configured to club, crush, flatten or otherwise damage the targeted pin or mushroom by generalized pressure exertion on the outer surface (and subsequently inner volume), in order to kill the targeted pin or mushroom. The blunt tool preferably comprises one or more effectively flat or low-curvature (blunt) surfaces which are directable to the targeted pin or mushroom. These surfaces have an area comparable to the pins preferably, in order to ensure that only the pins are destroyed without collateral damage.

[0208] Preferably, the blunt tool is configured to destroy the pin or mushroom by repetitive movements up and down and / or laterally, as such essentially hammering the pins or mushrooms and thereby destroying them.

[0209] Preferably, the blunt tool would be moved via translation via a robotic arm or via a rail and lowered to the desired position. Preferably, the angle of approach is directly vertically downward, although other angles of approach are possible, or even a combination of multiple angles, via multiple blunt tools.

[0210] In a preferred embodiment, the pin removal tool comprises one or more hot airflow tools configured to generate a (high pressure) heated, preferably unidirectional air flow, in order to kill the targeted pin or mushroom by drying it out. Alternatively, the air flow is not unidirectional. The hot air flow tool is in communication with a hot air source.

[0211] Preferably the hot air flow has a temperature of at least 50°C, preferably at least 60°C, more preferably at least 75°C, even more preferably at least 100°C, evenmore preferably at least 150°C, for instance at least 200°C, 250°C or more, such as 350°C or 500°C.

[0212] The tool may be configured for approaching or even contacting the surface of the targeted pin or mushroom and act upon the contacted surface, destroying the pin or mushroom via a hot air flow that is generated aimed at the pin or mushroom. The hot air flow may be generated at a high pressure.

[0213] In some embodiments, the tool may be inserted in the targeted pin or mushroom to act internally, and generate the hot air flow on the inside of the pin or mushroom from the inside, again destroying it by drying it out internally. In such a case, the hot air flow need not be unidirectional, and the tool can have a plurality of openings through which multiple separate air flows are provided.

[0214] The hot air flow tool preferably has a head with one or more openings through which the hot air flow can be generated and is in communication with a hot (high pressure) airflow generation means capable of generating a hot airflow to the tool and through the one or more openings.

[0215] Preferably, the hot air flow tool would be moved via translation via a robotic arm or via a rail and lowered to the desired position. Preferably, the angle of approach is directly vertically downward, although other angles of approach are possible, or even a combination of multiple angles, via multiple such tools.

[0216] In a preferred embodiment, the pin removal tool comprises one or more high water pressure tools configured to generate a high pressure, preferably unidirectional water flow, in order to kill the targeted pin or mushroom. The high water pressure tool is in communication with a high water pressure source.

[0217] In some embodiments, the water may be provided at high temperature, for instance at least 40°C, preferably at least 50°C, more preferably at least 60°C, even more preferably at least 70°C or even at least 80 or 90°C.

[0218] The high water pressure tool may be configured for approaching or even contacting the surface of the targeted pin or mushroom and act upon the contacted surface, destroying the pin or mushroom via a high water pressure that is generated aimed at the pin or mushroom. Alternatively, the high water pressure tool is used from a distance from the targeted pin or mushroom. The distance can vary, but is preferablyat least 2 mm, more preferably at least 5 mm, even more preferably at least 1.0 cm, even more preferably at least 2.5 cm, even more preferably at least 5 cm, even more preferably at least 10 cm. Preferably, said distance is at most 2 m, more preferably at most 1 m, even more preferably at most 50 cm, even more preferably at most 20 cm, even more preferably at most 10 cm.

[0219] In another alternative embodiments, the tool may be inserted in the targeted pin or mushroom to act internally, and generate the high pressure water flow on the inside of the pin or mushroom from the inside. In such a case, the high pressure water flow need not be unidirectional, and the tool can have a plurality of openings through which multiple separate water flows are provided.

[0220] The high water pressure flow tool preferably has a head with one or more openings through which the high pressure water flow can be generated and is in communication with a pressure means capable of generating a high pressure water flow to the tool and through the one or more openings.

[0221] Preferably, the high water pressure tool would be moved via translation via a robotic arm or via a rail and optionally lowered to the desired position. Preferably, the angle of approach is directly vertically downward, although other angles of approach are possible, or even a combination of multiple angles, via multiple such tools.

[0222] Alternatively or additionally, the high water pressure tool can be aimed at the desired target via a rotation of the tool, as this allows the tool to cover a wider area when using it to target pins or mushroom from a distance.

[0223] In any of the above aspects of the present invention, the mushrooms / pins / mycelium (hyphae) are preferably detected via vision technology by an algorithm specifically trained for these purposes. The algorithm receives as input image data that may or may not be preprocessed.

[0224] Preprocessing can involve a number of actions to augment the image quality, in terms of improving detection. These actions may include one or more of the following: blurring, scaling, color space manipulation, binarization, random drop-out, cropping, noise reduction, contrast enhancement.

[0225] The algorithm itself may use one or more of the following features to serve its functionality of detecting a desired object: upscaling, contour detection, edge detection, segmentation, pixel wise regression, feature extraction, depth estimation,anomaly detection, super resolution, instance segmentation, contrast adjustment, image stitching, temporal data analysis, etc.

[0226] The algorithm may use one or more of the following as potential bases: rule-based systems, foundation models, SAM (Segment Anything Model), FOMO (Faster Objects, More Objects), Lasagne, Tiramisu, YOLO (You Only Look Once), or can be trained from scratch.

[0227] The algorithm may run locally, at a processor on the premises, or can run on a remote server, or can run entirely or partly in the cloud.

[0228] In a most preferred version of the inventions, two or more of the aspects are combined with each other. It is clear to the skilled person reading this document that each separate aspect would synergistically augment the others by incorporating multiple aspect into a single mushroom cultivation method and associated system. For instance, improvements at the mycelium growth stage, by an optimized detection on when to act to slow down growth, trigger nodulation or other actions, would of course positively impact the further stages. It would however also further create the need for optimization down the line. As such, the combinations of two or more of the disclosed aspects are effectively disclosed in this document.

[0229] The present invention will be now described in more details, referring to examples that are not limitative.

[0230] EXAMPLES AND / OR DESCRIPTION OF FIGURES

[0231] With as a goal illustrating better the properties of the invention the following presents, as an example and limiting in no way other potential applications, a description follows of a number of preferred applications of the method for optimizing the mushroom cultivation, growing and harvesting process based on the invention.

[0232] In a first example, the method monitors a number of parameters at the growth bed wherein mycelium is growing. These parameters, amongst others, include the relative coverage of the growth bed with mycelium (coverage ratio), which is visually discernible due to the color differences, amongst others, as well as the shape (stringiness, etc.) of the mycelium.The cultivator can set a desired growth rate of this relative coverage overtime, with particular goals. In this case, the cultivator has set that the relative coverage should increase from 30% to 50% over a span of at most 2 hours.

[0233] As the system is monitoring this parameter carefully, it will detect 2 hours past the time at which the relative coverage has reached 30% whether or not the goal of 50% has been reached.

[0234] If this is not the case, the method can provide one or more corrective actions, either directly, such as modifying particular environmental parameters, or indirectly, such as alerting a cultivator and optionally suggest a direct action to be taken, for instance, modifying particular environmental parameters. This corrective action can furthermore be dependent on the amount with which the goal is missed.

[0235] In this example, should the relative coverage only reach 45% at the 2 hour mark, then this can result in an automatic temperature adjustment (increasing temperature) to speed up the mycelium growth.

[0236] The second example is a variation on the first example, but where the goal of 50% after 2 hours is overshot, and instead, a relative coverage of 70% is detected. In this case, the mycelium has expanded faster than expected, and again, the method can again provide for corrective actions. In this case, since the mycelium growth was faster than expected, a venting action (to jumpstart pin formation) and / or a temperature reduction (to slow down the mycelium growth) can be proposed (or enacted).

[0237] Additionally, the method can further be configured with a separate threshold for the two actions mentioned above (venting and temperature reduction), for instance, that venting is initiated (proposed) at 65% relative coverage, and that the temperature reduction is initiated (proposed) if the growth in 2 hours after reaching 30% relative coverage exceeds 50%.

[0238] In a further example, the mushroom size is tracked as a particular parameter, with the growth rate of the 75thpercentile (P75) being monitored in particular. Again, a preset value can be set by the cultivator as a goal, with this value potentially being set separately for specific time frames (for instance, hours 0 - X: growth rate goal A%; hours X - Y: growth rate goal B%; hours Y - Z: growth rate goal C%; or even in a dedicated function). In this case, the growth rate goal for the 75thpercentile is set at 4% / hour.

[0239] In reality, the growth rate for P75 is 6% / hour, meaning that the growth is happening much faster than expected. As a result, the method takes corrective actions (directly or indirectly), and proposes / enacts an earlier harvest timing, scheduling theharvesting an hour earlier than was originally the case, and / or proposes / enacts a temperature decrease of 1°C.

[0240] In a variation on the above example, the 50thpercentile (P50) is tracked, and has a preset growth rate goal of 4% / hour.

[0241] In reality, the growth rate goal is 2% / hour, lower than expected. In turn, the method takes corrective actions (directly or indirectly), and proposes / enacts a postponement of the harvest timing, scheduling the harvesting an hour later than originally the case, and / or proposes / enacts a temperature increase of 1°C, and / or proposes / enacts the provision of (additional) water.

[0242] Figure 1A and IB show a perspective view of a monitoring system (2) for performing the method, in the form of one or more cameras that capture (part of) the growth bed (4) below, holding soil and mycelium / pins / mushrooms (1). The monitoring system is mounted on a rail (2) spanning along the length of the growth bed (4), allowing either multiple of such monitoring systems (2) to be mounted in fixed positions along the length, in order to be able to capture (substantially) the entire growth bed (4), or to provide for one or more monitoring systems (2) which periodically move along the rail (2), again allowing (substantially) the entire growth bed (4) to be monitored.

[0243] The rail (2) provides for an easy way to provide power to the monitoring systems, as well as allowing communication between the monitoring system and an external system (that can in turn upload the data collected by the monitoring system to a server, and / or process the collected data locally).

[0244] Figure 2 shows the monitored / tracked parameters and environmental parameters over time (from 12 August llh30 to 20 August 19h30). In particular, it shows the temperature of the soil (compost) as line A (top most at the left side), the temperature of the air as line B (middle at the left side), and the mycelium coverage ratio as line C (bottom most at the left side).

[0245] It is distinctly visible that, as the air temperature is increased, which is the only parameter mentioned above that is directly controllable, around 15 August 19h30, triggers the mycelium coverage ratio to skyrocket.

[0246] Figure 3 shows a processed image from a mushroom growth bed, in which individual mushrooms (8) are identified via segmentation, with a contour (9) being provided onto the original image. This individual approach allows the tracking of individualized parameters as well as averages for (a part of) the population.Figure 4 shows an exterior perspective of a vertical cultivation system with multiple growth beds (4) holding soil (1) with mycelium, pinheads and / or mushrooms, with the growth beds (4) stacked on top of each other. Above each level of growth beds (4), a monitoring system (2) is provided with a field of view (10) on part of the growth bed. In this case, not the entire growth bed (4) is monitored. This can be remedied by providing movable monitoring systems (2), or increasing the number of monitoring systems (2) until a full(er) coverage is ensured.

[0247] Figure 5 shows a more detailed view of a level in the cultivation system of Figure 4, with the monitoring systems (2) mounted below the overlying level, above the growth bed (4) below, and with a field of view (10) that covers part of the soil (1) in the growth bed.

[0248] The present invention is in no way limited to the embodiments described in the examples and / or shown in the figures. On the contrary, methods according to the present invention may be realized in many different ways without departing from the scope of the invention. For instance, the method and system may be implemented in other types of cultivation, such as vegetables, fruit, and specifically spices and other more delicate types of produce.

Claims

43CLAIMS1. Computer-implemented method for monitoring mycelium, pin and mushroom growth in a mushroom bed via machine vision recognition, comprising the following steps:a. acquiring image data of the mushroom bed, wherein the image data is acquired along a line of sight substantially perpendicular to a plane defined by the mushroom bed;b. individually detecting a plurality of pins and / or mushrooms in the image data and registering the position of each of said detected pins and / or mushrooms, wherein the individual detection of pins is executed with a first machine learning model, and wherein the individual detection of mushrooms is performed with a second machine learning model different from the first machine learning model;c. determining one or more parameters associated to each of the detected pins and / or mushrooms from the image data, at least one of said parameters being representative of size of the pins and / or mushrooms;d. tracking the one or more parameters for each of the individual pins and / or mushrooms over time;e. proposing and / or initiating a reaction based on an analysis of said tracked parameters.

2. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to the preceding claim 1, wherein global parameters are determined associated to one or more bounded areas in the mushroom bed, said bounded areas comprising a cluster of multiple of the pins and / or mushrooms, wherein said global parameters are tracked overtime, with said global parameters representing a state of said bounded areas, and wherein the analysis further takes into account the global parameters.

3. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 2, wherein one or more events are registered with event data and logged with a time stamp, and associated to the mushroom bed and / or to the bounded areas within the mushroom bed, and / or to the detected pins and / or mushrooms within the mushroom bed, said events relating to a discontinuation of hydration and / or44a venting or airing action and the associated start of nodulation, and wherein the analysis further takes the one or more events into account.

4. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to the preceding claim 3, wherein the reaction is based on a subset of the parameters and optionally global parameters within a time period after the time stamp of the event data, said time period having a maximal duration of 3 days, and said time period being at least 12 hours after the time stamp of the event data.

5. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 4, wherein the step of individually detecting the plurality of pins and / or mushrooms is performed by a machine learning recognition model, wherein said recognition model is trained by supervised learning and / or reinforcement learning, wherein the recognition model is trained on a labeled mushroom dataset comprising a plurality of mushroom images in a mushroom bed, preferably under substantially similar conditions with respect to resolution, lighting and angle with respect to said mushroom bed as the image data, wherein mushrooms and / or pins in said mushroom images are labeled;and wherein the recognition model is configured to generate a bounding region on the image data representative of the mushrooms and / or pins that are individually detected in said image data, preferably wherein the image data is processed for quality augmentation with one or more of: blurring, scaling, binarization, random drop-out, noise reduction, cropping, contrast enhancement, color space manipulation.

6. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 5, wherein said one or more parameters comprises an occupied area on at least a subarea of the mushroom bed, wherein said subarea represents at least part of the surface of the mushroom bed, and wherein the analysis comprises determining a total mushroom bed surface occupancy quantification for said subarea from said occupied area of the individual mushrooms and / or pins, said total mushroom bed surface occupancy quantification representing the relative surface of the subarea of the mushroom bed being occupied by the detected mushrooms and / or modules, and wherein the analysis comprises tracking the change of said total mushroom bed surface occupancy quantification over time.

457. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 6, comprising a step of differentiating mycelium from non-mycelium in the image data of the mushroom bed, and determining a coverage ratio of the mycelium for at least a subarea of the mushroom bed, wherein said subarea represents at least part of the surface of the mushroom bed, the coverage ratio representing the relative surface of the subarea being occupied by the differentiated mycelium in the subarea, and wherein the analysis comprises tracking the change of the coverage ratio over time.

8. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to the preceding claim 7, wherein said step of differentiating mycelium is performed based on color, stringiness and / or flakiness of the image data.

9. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 7 to 8, further comprising a step of predicting a future mycelium density or population in the subarea based on the differentiated mycelium in the subarea, wherein said step of predicting is performed by a machine learning mycelium growth prediction model, wherein said mycelium growth prediction model is trained by supervised learning and / or reinforcement learning, wherein the mycelium growth prediction model is trained on a mycelium dataset comprising multiple sets of time sequence mycelium images, wherein each set comprises temporally sequential images of an at least partly overlapping subarea in a mushroom bed during mycelium growth, wherein the mycelium in the temporally sequential images is differentiated with respect to non-mycelium.

10. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to the preceding claim 9, wherein said mycelium growth prediction model is configured for, based on the image data for a subarea of the mushroom bed, generating a predicted future mycelium density image and / or predicted future mycelium density characteristics for said subarea, preferably wherein said image data for the subarea is processed for quality augmentation with one or more of: blurring, scaling, binarization, random drop-out, noise reduction, cropping, contrast enhancement, color space manipulation.

11. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 10, wherein environmental parameters are periodically measured and associated to the image data, preferably wherein timestamps of image data and associated environmental parameters substantially align, wherein the analysis further takes into account said associated environmental parameters.

12. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 11, and according to claim 9, wherein environmental parameters are associated to the mycelium images of the mycelium dataset, wherein timestamps of the mycelium images and the associated environmental parameters substantially align, and wherein the mycelium growth prediction model is trained taking into account said associated environmental parameters, wherein the step of predicting the future mycelium density or population further takes into account the environmental parameters associated to the image data for which the mycelium growth prediction model predicts the future mycelium density or population.

13. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to the preceding claim 11 or 12, wherein said environmental parameters comprise at least one, preferably two, more preferably three, and most preferably all, of: humidity, temperature, CO2 level and air flow at the mushroom bed.

14. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 13, wherein the analysis comprises differentiating the individually detected pins and / or mushrooms into growing objects and non-growing objects, based on change over time of the at least one of said parameters being representative of size according to a predetermined threshold for said change, wherein information regarding said differentiated growing and / or non-growing objects is used for said reaction and / or presented to an authorized user.

15. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to the preceding claim 14, wherein differentiating the individually detected pins and / or mushrooms into growing objects and non-growing objects is performed based on the changes over time in a period of at most 48 hours, preferably at most 24 hours, and more preferably at most 12 hours, prior to the present.

16. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 15, wherein the first and / or second machine learning model comprise a U-Net type model.

17. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to the preceding claim 16, wherein the first and / or second machine learning model comprise a U-Net type model and a subsequent Mask Region-based Convolutional Neural Network (R-CNN) model, wherein the U- Net type model performs a pixel-by-pixel evaluation of the image data to determine a likelihood of a pin- or mushroom pixel, and wherein the Mask R- CNN model subsequently performs an individual pin or mushroom instance detection on the image data augmented with the pixel-by-pixel evaluation.

18. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 17, wherein an individual growth metric is determined associated to each pin and / or mushroom, wherein a growth map is generated based on the image data and wherein the pins and / or mushrooms are visually marked based on the individual growth metric, wherein a first type of visual marker is applied to pins and / or mushrooms of which the individual growth metric is above a predefined first threshold value, and wherein a second type of visual marker is applied to pins and / or mushrooms of which the individual growth metric is below a predefined second threshold value, the second threshold value being equal to or below the first threshold value, and wherein the individual growth metric is based on a time-based evolution of the parameter representing the size.

19. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 18, further comprising a step of predicting a suitable harvesting time for one or more of the mushrooms based on the tracked parameters.

20. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 19, further48comprising a step of alerting when a suitable harvesting time is applicable for one or more of the mushrooms based on the tracked parameters.

21. Computer-implemented method for monitoring mycelium, pin and mushroom growth according to any one of the preceding claims 1 to 20, wherein the image data comprises digitally encoded image files, preferably compressed image files, and wherein the image data is combined with sensor data by means of sensor fusion, said sensor data comprising one or more of: air temperature, compost temperature, air humidity and CO2 level, and wherein the analysis is performed based on the fused image data and sensor data.

22. Computer-implemented method according to claim 21, wherein the sensor fusion comprises associating the sensor data with the image data based on corresponding timestamps and / or spatial alignment, and integrating the sensor data as additional input features for the analysis and / or for one or more of the machine learning models.

Images