A cadmium low-adsorption potato cultivation system

Abstract

The present application relates to the technical field of intelligent agriculture, and particularly relates to a cadmium low-adsorption potato cultivation system, which comprises: a genotype-phenotype prediction module, which is used for receiving genotype data of potato germplasm resources, processing the genotype data by using a basic genotype-phenotype correlation model based on machine learning, and outputting cadmium adsorption potential rating data of the potato germplasm resources to a selection and screening decision module; the selection and screening decision module is used for outputting a selection scheme data containing optimal germplasm identification and cultivation parameters by using a multi-objective optimization algorithm; the selection and screening decision module is further used for iteratively optimizing the basic genotype-phenotype correlation model and / or the multi-objective optimization algorithm according to cadmium adsorption capacity evaluation result data, so as to form a closed-loop feedback. The present application adopts a basic model combined with online iterative optimization, realizes rapid prediction of cadmium adsorption potential of potatoes and full-process automation and remote management of the low-cadmium high-yield target, and significantly improves breeding efficiency and precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent agricultural technology, in particular to a low-cadmium-adsorption potato cultivation system. BACKGROUND

[0002] Cadmium (Cd) is a heavy metal element with high toxicity and easy to accumulate in organisms, which poses a significant threat to the growth and development of potatoes, quality and safety, and human health. In cadmium-contaminated soil, potato plants can absorb cadmium ions through the root system and transport them to edible parts such as tubers. Long-term consumption of potatoes with excessive cadmium can cause damage to the human kidney and skeleton, and increase the risk of cancer. In addition, cadmium stress can also inhibit photosynthesis in potatoes, reducing their biomass and yield. Therefore, cultivating low-cadmium-adsorption potato varieties is one of the most cost-effective strategies to ensure the quality and safety of agricultural products and reduce dietary cadmium exposure risks from the source.

[0003] Currently, the cultivation of low-cadmium-adsorption potato varieties mainly relies on two technical paths: variety screening and gene regulation. In the aspect of variety screening, existing researches mostly use potting experiments combined with laboratory detection methods, such as measuring the cadmium content in different organs of 20 potato varieties by graphite furnace atomic absorption spectrophotometry, and screening low-accumulation varieties such as D47 and D157. However, this process requires digestion and detection of a large number of samples, which is tedious and time-consuming. In terms of detection equipment, although an online cadmium separation and enrichment device has been developed, which captures cadmium elements through electric heating evaporation-moderate resistance discharge, achieving on-site detection within half an hour to an hour, this device still needs to be used with an atomic spectrometer, and if a cadmium adsorption capacity prediction model is to be built, a large amount of sample data needs to be accumulated for training. In the field of gene regulation, it has been found that overexpression of StWRKY6 gene can reduce cadmium accumulation in edible parts of potatoes, but its cultivation effect still needs to be verified in combination with cadmium adsorption capacity detection. This makes the traditional technology where genotypic analysis, environmental control, and detection and evaluation are independent of each other, unable to achieve data intercommunication and automatic regulation. For example, the greenhouse environmental parameters need to be set manually, and the detection data need to be manually entered and analyzed, which is difficult to meet the needs of large-scale breeding. Therefore, there is an urgent need for a low-cadmium-adsorption potato cultivation system that can integrate genotypic and phenotypic prediction, environmental control, growth monitoring, and cadmium adsorption evaluation, and achieve full-process automation and remote management. SUMMARY

[0004] To solve the above problems, the present application provides a low-cadmium-adsorption potato cultivation system, which uses a basic model combined with online iterative optimization to achieve rapid prediction of potato cadmium adsorption potential and full-process automation and remote management of the low-cadmium-high-yield target, significantly improving breeding efficiency and accuracy.

[0005] In order to achieve the above object, the technical scheme of the present application is as follows: A cadmium low-adsorption potato cultivation system comprises a genotype-phenotype prediction module, a breeding and screening decision module, an automatic cultivation execution module, and a cadmium adsorption evaluation module.

[0006] The genotype-phenotype prediction module is used to receive genotype data of potato germplasm resources from an external database or user input, process the data using a machine learning-based basic genotype-phenotype correlation model, predict and output cadmium adsorption potential rating data of the potato germplasm resources to the breeding and screening decision module.

[0007] The breeding and screening decision module is used to receive the cadmium adsorption potential rating data, develop and output breeding program data containing preferred germplasm identification and cultivation parameters to the automatic cultivation execution module using a multi-objective optimization algorithm.

[0008] The automatic cultivation execution module is used to receive the cultivation program data, drive a controllable environment cultivation device according to the parameters of the cultivation program data, perform automatic cultivation operations, and real-time collect and output cultivation environment data and plant growth data to the cadmium adsorption evaluation module.

[0009] The cadmium adsorption evaluation module is used to receive the cultivation environment data and plant growth data, receive plant tissue cadmium content detection data from an external detection device, process the data using a preset adsorption capacity comprehensive evaluation model, calculate and output cadmium adsorption capacity evaluation result data to the breeding and screening decision module.

[0010] The breeding and screening decision module is also used to receive the cadmium adsorption capacity evaluation result data, and iteratively optimize the basic genotype-phenotype correlation model and / or the multi-objective optimization algorithm according to the cadmium adsorption capacity evaluation result data, forming a closed-loop feedback.

[0011] Further, the genotype-phenotype prediction module comprises:

[0012] The data preprocessing unit is used to receive genotype data of potato germplasm resources from an external database or user input, and output standardized genotype feature data to the model prediction unit using data cleaning, normalization, and feature dimension reduction processing methods.

[0013] The model prediction unit is used to receive the genotype feature data, call a pre-trained basic genotype-phenotype correlation model for processing, and output cadmium adsorption potential rating data of the potato germplasm resources to the breeding and screening decision module.

[0014] Further, the genotype-phenotype prediction module further comprises:

[0015] A model optimization unit is configured to receive the cadmium adsorption capacity evaluation results and corresponding genotype data, update a basic genotype-phenotype correlation model using an online learning algorithm, and output optimized model parameters to a model prediction unit.

[0016] Further, the breeding and screening decision module includes:

[0017] A germplasm preliminary selection unit is configured to receive cadmium adsorption potential rating data, select a candidate germplasm set from a germplasm resource library according to a preset potential threshold, and output the set to a scheme generation unit.

[0018] A scheme generation unit is configured to receive the candidate germplasm set, use a multi-objective optimization algorithm, and develop an initial breeding scheme for the candidate germplasm set with the goal of low cadmium adsorption and high yield, and output initial breeding scheme data to a scheme adjustment unit.

[0019] A scheme adjustment unit is configured to receive the initial breeding scheme data, integrate historical data from cadmium adsorption capacity evaluation results, combine real-time performance of the candidate germplasm in the cultivation process, calibrate and optimize the scheme, and output final breeding scheme data to an automated cultivation execution module.

[0020] A dynamic screening unit is configured to sort and re-screen the candidate germplasm according to cadmium adsorption capacity evaluation results and growth data after each cultivation and evaluation cycle, eliminate substandard germplasm, and feed back preferred germplasm identifiers to the germplasm preliminary selection unit to start the next breeding cycle.

[0021] Further, the automated cultivation execution module includes:

[0022] An environment control unit is configured to receive cultivation scheme data and combine it into an environment parameter set, generate control instructions, and output them to temperature and humidity, light, nutrient solution composition, and concentration control equipment in the controllable environment cultivation device.

[0023] A growth monitoring unit is configured to receive raw environmental and image data collected by sensors arranged in the controllable environment cultivation device, use image recognition and data fusion processing, and output cultivation environment data and plant growth data to the cadmium adsorption evaluation module.

[0024] Further, the growth monitoring unit includes:

[0025] An image processing subunit is configured to receive plant raw image data collected by a multispectral camera, use a deep learning image segmentation algorithm, extract and output growth index data including plant height, leaf area index, and chlorophyll relative content estimated based on spectral reflectance to a data fusion subunit.

[0026] A data fusion subunit is configured to receive growth index data of plant height, leaf area index and relative chlorophyll content, and real-time environmental data collected by sensors in the controllable environment cultivation device, perform spatio-temporal alignment and fusion processing, and output time-series aligned cultivation environment data and plant growth data to the cadmium adsorption evaluation module.

[0027] Further, the cadmium adsorption evaluation module comprises:

[0028] A data receiving unit is configured to receive cultivation environment data and plant growth data, and receive plant tissue cadmium content detection data provided by an external detection device, perform verification and formatting, and output standardized cadmium content data to the comprehensive evaluation unit.

[0029] A comprehensive evaluation unit is configured to receive standardized cadmium content data and plant growth data, call a dynamic weighting adsorption capacity evaluation model, and output cadmium adsorption capacity evaluation result data to the breeding and screening decision module.

[0030] Further, the adsorption capacity evaluation model is:

[0031] Adsorption capacity score = (1 - enrichment coefficient) x a + (1 - transport coefficient) x b + biomass coefficient x g;

[0032] Wherein a, b, and g are weight coefficients, and a + b + g = 1, and the weights can be dynamically adjusted according to the iterative evaluation results.

[0033] Further, the system further comprises:

[0034] A user interaction and visualization module is configured to receive user query and instruction data, and receive process data and result data output by each module of the system, perform visualization rendering, and output graphical interface data to a display device.

[0035] Further, the system is deployed on a cloud platform, and the automatic cultivation execution module performs data transmission with remote cultivation devices and sensing equipment through an Internet of Things protocol, receives control instructions and returns collected data.

[0036] The above scheme has the following beneficial effects:

[0037] 1. The scheme, by constructing a closed-loop control mechanism of "prediction-decision-execution-evaluation-feedback", effectively overcomes the problems of traditional potato cadmium adsorption performance detection, such as dependence on a large amount of pre-training data, model update lag and long breeding cycle. The system can quickly output prediction results based on an initial basic model, and can collect data in real time during variety breeding and identification, dynamically optimize the model, and significantly improve the efficiency and accuracy of cadmium low-adsorption variety breeding.

[0038] 2、The scheme realizes intelligent and automatic regulation and control of the breeding process, can simultaneously consider low cadmium adsorption and high yield breeding goals according to a multi-objective optimization algorithm, and improves the synergy and practicality of the breeding scheme. By introducing a dynamic weighted adsorption capacity evaluation model, the system can flexibly adjust the evaluation index weight, and enhance the adaptability to different cultivation conditions and breeding needs.

[0039] 3、The scheme also supports cloud platform deployment and remote interaction, facilitates multi-terminal access and collaborative operation, and greatly improves the scalability and practical value of the system. Ultimately, the system not only realizes efficient and low-cost low-cadmium adsorption breeding of potatoes, but also provides a referenceable technical framework for heavy metal resistance breeding of other crops.

[0040] Additional aspects and advantages of the application will be described in the following description, some of which will become apparent to those skilled in the art from the following description, some of which will become apparent from the following description, or will be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 The system framework schematic diagram of the embodiment of the low-cadmium adsorption potato breeding system of the application. DETAILED DESCRIPTION

[0042] The technical solutions of the application will be described in detail below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0043] In the description of the application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore cannot be understood as a limitation on the application. In addition, the terms "first", "second", "third" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance.

[0044] In the description of the application, it should be noted that unless otherwise specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integrally connected; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through intermediate medium, or the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the application can be understood according to the specific circumstances.

[0045] The following is further described in detail through a specific embodiment: Embodiment:

[0046] In the existing potato variety breeding and cadmium pollution prevention and control research, the traditional method often relies on a large number of historical data to pre-train a complex prediction model, and then evaluates and selects the cadmium adsorption capacity. This method not only has high data acquisition cost and long model training period, but also is difficult to adapt to the dynamic changes of potato genotype-phenotype relationship in different regions and different cultivation conditions, resulting in low detection efficiency and response lag, and unable to realize rapid and accurate breeding of low cadmium adsorption varieties.

[0047] Based on the above problems, the inventors propose to build a closed-loop control mechanism of "prediction-decision-execution-evaluation-feedback" to realize the dynamic evaluation and iterative optimization of the cadmium adsorption capacity of potatoes. This system not only can quickly output prediction results based on the initial basic model, but also can continuously collect data and optimize the model during actual cultivation, significantly improving the efficiency and accuracy of low cadmium adsorption variety breeding.

[0048] A certain agricultural academy plans to breed a low-adsorption potato variety suitable for planting in cadmium-polluted soil. Due to the traditional method, thousands of samples of genotype and cadmium content data need to be collected first, and then a machine learning model is trained to predict new germplasm. The entire process takes several months, and the prediction results are greatly affected by the quality of the initial data. Therefore, a cadmium low-adsorption potato breeding system is proposed, as shown in FIG. 1, which includes a genotype-phenotype prediction module, a breeding and screening decision module, an automated cultivation execution module, and a cadmium adsorption evaluation module. Figure 1

[0049] The genotype-phenotype prediction module is used to receive genotype data of potato germplasm resources from external databases or user input, process it using a basic genotype-phenotype correlation model based on machine learning, and predict and output cadmium adsorption potential rating data of potato germplasm resources to the breeding and screening decision module. The genotype-phenotype prediction module includes:

[0050] The data preprocessing unit is used to receive genotype data of potato germplasm resources from external databases or user input, and uses data cleaning, normalization, and feature dimension reduction processing methods to output standardized genotype feature data to the model prediction unit. The following specific steps are used for processing:

[0051] S11, data cleaning: missing values in the original genotype data are filled using the mean imputation method, i.e. replaced by the genotype frequency mean of all samples at that SNP site; abnormal values exceeding ±3 standard deviations are considered as detection noise and are removed.

[0052] ​Normalization: To eliminate the influence of different dimensions, the minimum-maximum normalization method is used to linearly map the data of each SNP site to the interval [0, 1], and the calculation formula is:

[0053]

[0054] S12, where X is the original value of the data of each SNP site, Xmin and Xmax are the minimum and maximum values of all samples of the site, respectively;

[0055] S13, feature dimension reduction: To reduce data redundancy and highlight key features, principal component analysis (PCA) method is used. The first N principal components with cumulative variance contribution rate ≥95% are reserved to form standardized genotype feature data with unified dimensions and standard numerical ranges;

[0056] A model prediction unit is configured to receive the genotype feature data, call the pre-trained basic genotype-phenotype association model for processing, and output the cadmium adsorption potential rating data of the potato germplasm resources to the breeding and screening decision module. Specifically, the basic genotype-phenotype association model is a random forest classifier, and its pre-training process is as follows: the genotype feature data of at least 500 potato germplasm resources accumulated historically and their corresponding cadmium adsorption phenotype labels (divided into three levels of “high”, “medium” and “low”) are used as the training set. During training, the core parameters of the random forest are set as follows: the number of decision trees , the maximum depth of the decision tree , and the minimum number of samples required for the leaf node . After evaluating and optimizing the model performance through ten-fold cross-validation, the model parameters are saved for calling by the prediction unit. The feature data received by the unit is input into the model, and the cadmium adsorption potential rating data is output.

[0057] A model optimization unit is configured to receive the cadmium adsorption capacity evaluation result and the corresponding genotype data fed back by the cadmium adsorption evaluation module, update the basic genotype-phenotype correlation model by using an online learning algorithm, and output the optimized model parameters to the model prediction unit. Specifically, the online updating of the model is implemented by using an incremental random forest algorithm: after each breeding and evaluation cycle, the newly generated "genotype feature data-cadmium adsorption score" paired data is taken as an incremental batch. The system calls the incremental data in the basic genotype-phenotype correlation model database, re-trains part of the decision trees in the original random forest model using the incremental data, updates the splitting rules of the tree nodes, thereby realizing the iterative optimization of the model parameters without re-training the entire model. The optimized model parameters are output in real time to the model prediction unit for prediction in the next round. First, the researchers input the genotype data of the existing 300 potato germplasm resources into the system. The genotype-phenotype prediction module is started, and the data preprocessing unit of the genotype-phenotype prediction module cleans, normalizes and performs principal component analysis (PCA) dimension reduction processing on the original genotype data to extract the most representative feature data. Subsequently, the model prediction unit calls the pre-trained random forest model (basic genotype-phenotype correlation model) to output the cadmium adsorption potential rating (divided into three levels of high, medium and low) of each germplasm.

[0058] A breeding and screening decision module is configured to receive the cadmium adsorption potential rating data, develop and output a breeding scheme data including the identification of the preferred germplasm and the breeding parameters to the automatic breeding execution module by using a multi-objective optimization algorithm; wherein the breeding and screening decision module is further configured to receive the cadmium adsorption capacity evaluation result data, and iteratively optimize the basic genotype-phenotype correlation model and / or the multi-objective optimization algorithm according to the cadmium adsorption capacity evaluation result data to form a closed-loop feedback.

[0059] The breeding and screening decision module includes:

[0060] A germplasm preliminary selection unit is configured to receive the cadmium adsorption potential rating data, select a candidate germplasm set from the germplasm resource library according to a preset potential threshold, and output the set to a scheme generation unit.

[0061] A scheme generation unit is configured to receive the candidate germplasm set, develop and output initial breeding scheme data for the candidate germplasm set to a scheme adjustment unit by using a multi-objective optimization algorithm with low cadmium adsorption and high yield as the target. The specific steps of the multi-objective optimization algorithm are as follows:

[0062] S21, encoding and population initialization:

[0063] S211, input: the identifier of each germplasm in the candidate germplasm set and the corresponding cadmium adsorption potential rating data.

[0064] S212, coding method: real number coding, each individual (chromosome) represents a breeding scheme, which is composed of a set of breeding parameters, including: nutrient solution cadmium ion concentration (unit: μM, range: 0.1-5.0), photoperiod (unit: hours / day, range: 12-16), day-night temperature difference (unit: ℃, range: 8-12).

[0065] S213, population initialization: randomly generate an initial population within the allowed range of each parameter, and set the population size to 50.

[0066] S22, fitness evaluation:

[0067] S221, define two objective functions for double-objective optimization of low cadmium adsorption and high yield:

[0068] Objective function 1 (minimization): f1 = normalized cadmium adsorption potential score. This score comes directly from the output of the genotype-phenotype prediction module, which has been mapped to the [0, 1] interval, and a higher value indicates a greater cadmium adsorption potential, so it needs to be minimized.

[0069] Objective function 2 (maximization): f2 = yield potential index based on historical data. This index is obtained by querying the historical database to obtain the average tuber biomass of the germplasm or its parents under similar breeding conditions according to the germplasm identifier, and is normalized to the [0, 1] interval.

[0070] S222, the algorithm calculates the value of each individual on the two objective functions, performs fast non-dominated sorting, and determines the Pareto rank of the individual.

[0071] S23, selection, crossover and mutation:

[0072] S231, selection: binary tournament selection method is used, and selection is performed according to the non-dominated rank and crowding degree of the individual.

[0073] S232, crossover: simulated binary crossover is used, and the crossover probability is set to 0.8.

[0074] S233, mutation: polynomial mutation is used, and the mutation probability is set to 0.1.

[0075] S234, iteration: set the number of evolutionary generations to 100 generations.

[0076] S24, output:

[0077] S241, after the algorithm runs, select the Pareto frontier with the highest non-dominated rank from the final generation population.

[0078] S242, the system selects an individual from the front that performs best on the objective function f2 by default, decodes it into a specific breeding parameter combination, and outputs it as initial breeding scheme data;

[0079] The scheme adjustment unit receives the initial breeding scheme data, integrates historical data from the cadmium adsorption capacity evaluation results, combines the real-time performance of the candidate germplasm in the breeding process, performs scheme calibration and optimization, and outputs the final breeding scheme data to the automated breeding execution module. The steps of scheme calibration and optimization are as follows:

[0080] S31, data preparation:

[0081] S311, real-time performance data: continuously receive the daily average growth rate of leaf area index uploaded by the growth monitoring unit.

[0082] S312, historical data: retrieve the final cadmium adsorption capacity score of the current candidate germplasm set (or its parent) under different breeding parameters in the past breeding cycle from the system database.

[0083] S32, calibration trigger mechanism:

[0084] S321, the system sets a threshold for the daily average growth rate of leaf area index (for example: 0.08).

[0085] S322, if the system detects that the growth rate of a candidate germplasm is continuously lower than the threshold for 3 consecutive days, the breeding scheme calibration process for the germplasm is automatically triggered.

[0086] S33, calibration execution logic:

[0087] S331, the system establishes a simple linear regression model: ; wherein, represents the baseline predicted score when all breeding parameters are 0; represents the influence weight of nutrient solution cadmium concentration on the predicted score; the nutrient solution cadmium concentration is an input variable with units of μm, taken from the current breeding scheme or historical data; represents the influence weight of light duration on the predicted score; the light duration is an input variable with units of hours / day, taken from the current breeding scheme or historical data. The linear regression model is trained using the historical data obtained in S312.

[0088] S332, when calibration is triggered, the system applies this model to the current scheme:

[0089] If the current real-time growth performance is far below the expected value, and the model shows that reducing the cadmium concentration in the nutrient solution (for example: reduce 0.5 μM) has a negative impact on the cadmium adsorption score within an acceptable range (for example: the predicted decrease is not more than 5%), this adjustment is performed, giving priority to ensuring growth.

[0090] If the historical model shows that a small increase in light duration (for example: increase 0.5 hours) under the current parameters may be beneficial to growth and reduce cadmium adsorption, this adjustment is performed.

[0091] S333, the calibration process is based on rules and model predictions, and the adjustment of any parameter at a time does not exceed 10% of the initial value.

[0092] S34, output: integrate the cultivation parameters calibrated by the above logic into the final breeding scheme data and output;

[0093] Dynamic screening unit, for sorting and re-screening candidate germplasm after each round of cultivation and evaluation period, eliminating substandard germplasm, and feeding back the preferred germplasm identification to the germplasm preliminary selection unit for starting the next round of breeding cycle.

[0094] The rating data is input into the breeding and screening decision module. First, the germplasm preliminary selection unit screens the candidate germplasm set (for example, the preferred 10 germplasm numbers) from all germplasms according to the preset potential threshold. Then, the scheme generation unit uses a multi-objective genetic algorithm to develop an initial breeding scheme for the candidate germplasm set, including corresponding cultivation parameters (such as nutrient solution Cd²⁺ concentration, pH value, light cycle, etc.). Through further fine-tuning by the scheme adjustment unit combined with historical evaluation results, the final breeding scheme is formed and output to the automated cultivation execution module. After each round of cultivation and evaluation period, the dynamic screening unit will sort and re-screen the candidate germplasm according to the cadmium adsorption capacity evaluation results and growth data, and the winners and losers will be selected, and the preferred germplasm identification will be fed back to the system to start the next round of progressive breeding cycle.

[0095] Automatic cultivation execution module, for receiving cultivation scheme data, driving controllable environment cultivation device according to the parameters of cultivation scheme data, executing automatic cultivation operation, and collecting and outputting cultivation environment data and plant growth data to cadmium adsorption evaluation module in real time; the automatic cultivation execution module includes:

[0096] Environment control unit, for receiving cultivation scheme data and collecting into environment parameter set, generating control instructions, and outputting to temperature and humidity, light, nutrient solution composition and concentration control equipment in controllable environment cultivation device.

[0097] Specifically, the unit communicates with the remote cultivation device through a signal connection and instructs the cultivation device to generate and issue instructions:

[0098] Temperature control: Set the target temperature Ttarget (e.g., 22°C). The unit reads the real-time data Tcurrent from the temperature sensor and uses a PID control algorithm (proportional coefficient Kp = 2.0, integral time Ti = 300 seconds, and differential time Td = 50 seconds) to calculate the control amount. The power of the heater or refrigerator is adjusted through a PWM signal to maintain the temperature and humidity within ±2°C of Ttarget.

[0099] Nutrient solution control: According to the target Cd²⁺ concentration Ctarget in the scheme, the unit controls the peristaltic pump to inject Cd²⁺-containing nutrient mother liquor into the mother liquor pool at a specific speed. At the same time, the pH value of the nutrient solution is monitored in real time through the pH sensor, and if it exceeds the set range (e.g., 5.5-6.5), dilute acid or dilute base solution is automatically added for adjustment.

[0100] Light control: According to the light cycle set in the scheme, the unit controls the power of the LED supplemental light group through a relay switch to achieve precise timing switching. The light intensity is adjusted by adjusting the LED driving current to achieve stepless control.

[0101] The growth monitoring unit is used to receive raw environmental and image data collected from sensors arranged in the controllable environment cultivation device. It uses image recognition and data fusion processing to output cultivation environment data and plant growth data to the cadmium adsorption evaluation module. The growth monitoring unit includes:

[0102] The image processing subunit is used to receive raw plant image data collected by the camera. It uses a deep learning image segmentation algorithm to extract and output plant height and leaf area index growth indicator data to the data fusion subunit. Specifically, the image processing subunit receives raw plant image data collected periodically (e.g., every day at 10:00 am) by a multispectral camera deployed above the cultivation area. A deep learning model based on the U-Net architecture is used for image segmentation. The model is pre-trained using 1000 potato plant images labeled with the LabelMe tool. The input image size is 256x256 pixels, and the output is a plant pixel-level segmentation mask. When extracting features from the image:

[0103] Plant height: According to the segmented plant mask, calculate the vertical distance between the highest pixel point and the pixel point of the cultivation substrate base surface, and convert it to the actual physical height (centimeters) based on the camera calibration parameters.

[0104] Leaf area index: Calculate the total pixel area of the plant mask and convert it according to the pre-labeled relationship model between pixel area and actual leaf area.

[0105] The extracted plant height, leaf area index and other growth index data are packaged and output to the data fusion subunit.

[0106] The data fusion subunit is used for receiving plant height and leaf area index growth index data, and real-time environmental data collected by sensors in the controllable environment cultivation device, performing spatio-temporal alignment and fusion processing, and outputting time-series aligned cultivation environment data and plant growth data to the cadmium adsorption evaluation module. By receiving growth index data from the image processing subunit and real-time environmental data (temperature, humidity, light intensity) from environmental sensors, the Kalman filter algorithm is used for spatio-temporal alignment and fusion of multi-source data. The system maintains a state vector [plant height, leaf area index, temperature, humidity] and performs optimal estimation of image data (low frequency but rich in information) and environmental sensor data (high frequency but single) every minute to eliminate transient noise, output time-series aligned and data consistent cultivation environment data and plant growth data, and output to the cadmium adsorption evaluation module in JSON format.

[0107] After the automatic cultivation execution module receives the scheme, the environmental control unit drives the controllable environment cultivation device (such as an intelligent greenhouse) to adjust temperature and humidity, light intensity, nutrient solution composition and concentration. At the same time, the growth monitoring unit collects plant images and environmental data in real time through the deployed multi-spectral camera and environmental sensors. The image processing subunit uses the U-Net deep learning model to segment the plant image and extract growth indicators such as plant height and leaf area index; the data fusion subunit performs spatio-temporal alignment and fusion of image data and environmental sensor data to form a time-series consistent cultivation dataset, which is transmitted to the cadmium adsorption evaluation module.

[0108] The cadmium adsorption evaluation module is used for receiving cultivation environment data and plant growth data, and receiving plant tissue cadmium content detection data from external detection devices, processing using a preset adsorption capacity comprehensive evaluation model, calculating and outputting cadmium adsorption capacity evaluation result data to the selection and screening decision module; the cadmium adsorption evaluation module includes:

[0109] The data receiving unit is used for receiving cultivation environment data and plant growth data, and receiving plant tissue cadmium content detection data provided by external detection devices, performing verification and formatting, and outputting standardized cadmium content data to the comprehensive evaluation unit. Specifically, the external detection device (such as an atomic absorption spectrometer) sends detection data to the data receiving unit through the RS-232 serial port, and the data receiving unit performs CRC32 verification on the received data to ensure data transmission integrity; the cadmium content data is inversely calculated according to the standard curve used during detection, and the unit is unified as mg / kg (based on dry weight), and the standardized cadmium content data is output to the comprehensive evaluation unit.

[0110] The comprehensive evaluation unit is configured to receive the standardized cadmium content data and the plant growth data, call the dynamic weighting adsorption capacity evaluation model, and output cadmium adsorption capacity evaluation result data. The formula of the adsorption capacity evaluation model is:

[0111] Adsorption capacity score = (1 - enrichment coefficient) x a + (1 - transport coefficient) x b + biomass coefficient x g

[0112] wherein the enrichment coefficient = tuber cadmium content / soil cadmium content; the soil cadmium content is a known value set at the initial stage of cultivation.

[0113] The transport coefficient = tuber cadmium content / root system cadmium content; the root system cadmium content and the tuber cadmium content are measured at the same period.

[0114] The biomass coefficient = fresh weight of tubers per plant / maximum fresh weight of tubers per plant of all candidate germplasms in the current cultivation round.

[0115] wherein a, b, and g are weight coefficients, and a + b + g = 1.

[0116] The weights can be dynamically adjusted according to the iteration evaluation results, and the cadmium adsorption capacity evaluation result data is output to the breeding and screening decision module. The dynamic adjustment method of the weights is as follows: after each round of evaluation, the unit records the comprehensive score under the current weights, and compares it with an "ideal target" based on the actual cadmium content and yield (for example, setting tuber cadmium content <= 0.2 mg / kg and biomass >= 200 g / plant as the ideal state). The gradient descent method is used for fine tuning: calculate the error between the comprehensive score and the ideal state, then adjust a, b, and g in the direction of error reduction with a learning rate of 0.05. After adjustment, the weights are normalized to ensure that their sum is always 1. For example, if it is found in a certain iteration that the score of a germplasm with high biomass but also high cadmium adsorption is too high, the system will automatically increase the weights of a and b and reduce the weight of g in the next round, so that the evaluation standard is more inclined to "low cadmium adsorption".

[0117] When the potatoes grow to the tuber formation stage, the researchers take samples for testing, obtain plant tissue cadmium content data, and input the system. The data receiving unit of the cadmium adsorption evaluation module performs verification and standardization processing on the test data, and the comprehensive evaluation unit calls the dynamic weighting adsorption capacity evaluation model, wherein a, b, and g are weight coefficients, initially set to 0.4, 0.3, and 0.3, and dynamically adjusted according to the previous evaluation results. The evaluation results of this batch show that the cadmium adsorption scores of 3 germplasms are significantly better than those of the remaining germplasms.

[0118] The evaluation results are fed back to the selection and screening decision module, triggering the model optimization unit to update the basic genotype-phenotype correlation model using an online learning algorithm (such as incremental random forest). At the same time, the multi-objective optimization algorithm also adjusts the weight settings according to the latest evaluation results to generate more accurate next round breeding schemes.

[0119] The system also includes:

[0120] A user interaction and visualization module is used to receive user queries and instruction data, and to receive process data and result data output by the various modules of the system, for visualization rendering and output of graphical interface data to a display device. The system is deployed on a cloud platform, and the automated breeding execution module performs data transmission with remote breeding devices and sensing equipment through an Internet of Things protocol, receives control instructions, and transmits back collected data.

[0121] The user interaction and visualization module provides a graphical interface throughout the process, supporting researchers to query module data in real time, adjust system parameters, and perform backtracking and analysis of the breeding process.

[0122] Compared to traditional methods, the system quickly screens out low-cadmium potential germplasm in the first round of breeding, and continuously optimizes the model and scheme in subsequent iterations, significantly shortening the breeding cycle and reducing experimental costs. The system realizes model self-evolution through a closed-loop feedback mechanism, avoiding excessive dependence on initial data volume, and is particularly suitable for actual breeding scenarios with limited resources and variable environments.

[0123] In addition, the system is deployed on a cloud platform, supporting multi-terminal remote access and control, and realizing data interconnection with remote breeding equipment through an Internet of Things protocol, with good scalability and practicality.

[0124] I. Experimental design background and purpose

[0125] Traditional low-cadmium adsorption potato breeding techniques have problems such as independent analysis of genotypes, environmental control, and detection and evaluation, data cannot be interconnected, and lack of automated control, resulting in low breeding efficiency. This experiment compares the new breeding system with traditional methods to verify the advantages of the new breeding system in full-process automation, data interconnectivity, and breeding efficiency.

[0126] II. Experimental grouping and material preparation

[0127] Test group: Use the cadmium low adsorption potato breeding system of the invention, including the genotype-phenotype prediction module, the selection and screening decision module, the automated breeding execution module, and the cadmium adsorption evaluation module. The system is deployed on a cloud platform and connected to a remote intelligent greenhouse.

[0128] Control group: The traditional pot experiment combined with laboratory testing methods was used, including artificial genotype screening, manual setting of greenhouse environmental parameters (temperature, humidity, light, etc.), manual collection of growth data and cadmium content detection.

[0129] Experimental materials: 300 potato germplasm resources (genotype data known) were selected and divided into two groups of 150 each. Each group included known low cadmium adsorption varieties (such as D47 and D157) and common varieties.

[0130] III. Experimental Procedure

[0131] (I) Experimental Group Procedure

[0132] Genotype-phenotype prediction

[0133] The data preprocessing unit cleaned, normalized, and performed PCA dimensionality reduction on 150 germplasm genotype data.

[0134] The model prediction unit calls a pre-trained random forest model to output a cadmium adsorption potential rating (high / medium / low), and selects the top 30 potential germplasm samples.

[0135] Cultivation Program Development

[0136] The breeding and screening decision module uses a multi-objective genetic algorithm to generate an initial breeding plan (including parameters such as cadmium concentration in nutrient solution, pH value, and photoperiod) with the goals of "low cadmium adsorption" and "high yield".

[0137] The scheme adjustment unit optimizes the scheme by combining historical evaluation data and outputs it to the automated cultivation execution module.

[0138] Automated cultivation and monitoring

[0139] The environmental control unit drives the intelligent greenhouse to automatically adjust the temperature and humidity (22±2℃, humidity 60±5%), light intensity (16h photocycle, 300μmol / m²・s) and nutrient solution composition.

[0140] The growth monitoring unit collects plant images and environmental data in real time through multispectral cameras and sensors. The image processing subunit extracts indicators such as plant height and leaf area index, and the data fusion subunit generates a time-series dataset.

[0141] Cadmium adsorption evaluation and model iteration

[0142] Samples were taken during the tuber formation period, and the cadmium content of the plant tissue was measured by an external detection device. The cadmium adsorption evaluation module calculated the adsorption capacity score through a dynamic weighted model (formula: adsorption capacity score = (1 - enrichment coefficient) × 0.4 + (1 - translocation coefficient) × 0.3 + biomass coefficient × 0.3).

[0143] The breeding and screening decision module iteratively optimizes the genotype-phenotype correlation model and the multi-objective algorithm according to the evaluation results, and generates a next round of scheme.

[0144] (B) Control group process

[0145] Genotype screening: manual comparison of literature data, screening of 150 germplasm possible low cadmium varieties, time-consuming 3 days.

[0146] Environmental control: manual setting of greenhouse parameters (same as the test group), manual recording of temperature and humidity every day, error ±3℃ / ±10%.

[0147] Growth data collection: manual measurement of plant height and leaf area every 7 days, manual recording of data, time-consuming about 2 hours / time.

[0148] Cadmium content detection: sampling at the tuber formation stage, manual digestion, and detection using graphite furnace atomic absorption spectrophotometry, each sample detection time-consuming 4 hours, data manual entry analysis.

[0149] Four, detection target data

[0150] Breeding cycle: the total time from genotype screening to obtaining low cadmium varieties.

[0151] Labor cost: manual operation time (hours) required throughout the experiment.

[0152] Data interoperability: automatic data transmission rate (%) of each link, traditional method is 0%, and the system is 100%.

[0153] Cadmium adsorption capacity: cadmium content in tubers (mg / kg), enrichment coefficient, transport coefficient and biomass (g / plant).

[0154] Model iteration efficiency: the accuracy rate of the model prediction is improved after each round of breeding.

[0155] Five, experimental results table

[0156] Table 1 Comparison of experimental results between test group and control group

[0157]

[0158] Six, result discussion

[0159] The experimental data show that the cadmium low adsorption potato breeding system of the present application is significantly better than the traditional method in many aspects:

[0160] 1. Significant improvement in breeding efficiency: The test group completed three rounds of screening and evaluation within 180 days through systematic iterative optimization, while the traditional method required 210 days for a single round of screening. Although the total period is similar, the system achieves multiple rounds of optimization in the same time, significantly improving the density and efficiency of the breeding process. In terms of labor costs, the test group is only 120 hours, which is 42.9% lower than the 210 hours of the control group, reflecting the significant advantages of the system in automated management and data integration.

[0161] 2. Full-process data intercommunication and automation: The system realizes full-process data automatic transmission and processing from genotype prediction, decision making, environment control to cadmium adsorption evaluation, with a transmission rate of 100%. In contrast, the traditional method relies on manual recording and input, which is not only inefficient, but also has the risk of data inconsistency and human error.

[0162] 3. Good cadmium adsorption control and yield balance: The test group's tuber cadmium content is 0.15 mg / kg, significantly lower than the control group's 0.25 mg / kg, with a decrease of 40%. The enrichment coefficient and transport coefficient decreased by 36.2% and 37.5%, respectively, indicating that the system has obvious effect in inhibiting cadmium absorption and transport. At the same time, the biomass of the test group increased by 13.6%, indicating that the system has not sacrificed yield while achieving the goal of "low cadmium", reflecting the effectiveness and practicality of the multi-objective optimization algorithm.

[0163] 4. Model has the ability to continuously optimize: The system improves the model prediction accuracy from 70% to 82% in three iterations through online learning mechanism, with an increase of 17%. In contrast, the traditional method relies on static models and cannot dynamically adjust according to actual cultivation data, with a fixed prediction accuracy of 65%. This shows that the system has good self-adaptation and evolution ability, especially suitable for complex and variable breeding environment.

[0164] In summary, the system successfully solves the problem of independent links and lack of automation in traditional technology, providing an efficient and accurate technical solution for cadmium low adsorption potato breeding, and also providing a reference framework for heavy metal resistance breeding of other crops.

[0165] Obviously, the above embodiments are only examples for clarity and do not limit the implementation. For ordinary skilled persons in the art, other different forms of changes or variations can be made based on the above description. Here, it is not necessary or possible to exhaust all implementation methods. The obvious changes or variations derived therefrom are still within the protection scope of the present invention.

Claims

1. A cadmium-low adsorption potato cultivation system, characterized in that, The system comprises a genotype-phenotype prediction module, a breeding and screening decision module, an automated cultivation execution module, and a cadmium adsorption evaluation module. The genotype-phenotype prediction module is configured to receive genotype data of potato germplasm resources from an external database or user input, process the genotype data by using a machine learning-based basic genotype-phenotype correlation model, predict and output cadmium adsorption potential rating data of the potato germplasm resources to the breeding and screening decision module. The breeding and screening decision module is configured to receive the cadmium adsorption potential rating data, develop and output a breeding scheme data including preferred germplasm identification and cultivation parameters to the automated cultivation execution module by using a multi-objective optimization algorithm. The automated cultivation execution module is configured to receive the breeding scheme data, drive a controllable environment cultivation device according to parameters of the breeding scheme data, perform automated cultivation operations, and collect and output cultivation environment data and plant growth data to the cadmium adsorption evaluation module in real time. The cadmium adsorption evaluation module is configured to receive the cultivation environment data and the plant growth data, receive plant tissue cadmium content detection data from an external detection device, process the data by using a preset adsorption capacity comprehensive evaluation model, calculate and output cadmium adsorption capacity evaluation result data to the breeding and screening decision module. The breeding and screening decision module is further configured to receive the cadmium adsorption capacity evaluation result data, and iteratively optimize the basic genotype-phenotype correlation model and / or the multi-objective optimization algorithm according to the cadmium adsorption capacity evaluation result data to form a closed-loop feedback. The breeding and screening decision module comprises: A germplasm preliminary selection unit configured to receive the cadmium adsorption potential rating data, select a candidate germplasm set from a germplasm resource library according to a preset potential threshold, and output the set to a scheme generation unit. The scheme generation unit is configured to receive the candidate germplasm set, develop and output initial breeding scheme data for the candidate germplasm set to a scheme adjustment unit by using a multi-objective optimization algorithm with low cadmium adsorption and high yield as the target. The scheme adjustment unit is configured to receive the initial breeding scheme data, integrate historical data from cadmium adsorption capacity evaluation results, combine real-time performance of the candidate germplasm in the cultivation process, perform scheme calibration and optimization, and output final breeding scheme data to the automated cultivation execution module. A dynamic screening unit is configured to sort and re-screen the candidate germplasm according to the cadmium adsorption capacity evaluation results and growth data after each cultivation and evaluation cycle, eliminate substandard germplasm, and feed back preferred germplasm identification to the germplasm preliminary selection unit to start the next breeding cycle. The cadmium adsorption evaluation module comprises: A data receiving unit configured to receive the cultivation environment data and the plant growth data, receive plant tissue cadmium content detection data provided by an external detection device, perform verification and formatting, and output standardized cadmium content data to a comprehensive evaluation unit. The comprehensive evaluation unit is configured to receive the standardized cadmium content data and the plant growth data, call a dynamic weighting adsorption capacity evaluation model, and output cadmium adsorption capacity evaluation result data to the breeding and screening decision module.

2. The cadmium low-adsorbing potato breeding system according to claim 1, characterized by The genotype-phenotype prediction module comprises: A data preprocessing unit is configured to receive genotype data of potato germplasm resources from an external database or user input, adopt data cleaning, normalization and feature dimension reduction processing, and output standardized genotype feature data to a model prediction unit. The model prediction unit is configured to receive the genotype feature data, call a pre-trained basic genotype-phenotype correlation model for processing, and output cadmium adsorption potential rating data of the potato germplasm resources to a breeding and screening decision module.

3. The cadmium low-adsorbing potato breeding system according to claim 2, characterized by The genotype-phenotype prediction module further comprises: A model optimization unit is configured to receive cadmium adsorption capacity evaluation results and corresponding genotype data, update the basic genotype-phenotype correlation model using an online learning algorithm, and output optimized model parameters to the model prediction unit.

4. The cadmium low-adsorbing potato breeding system according to claim 3, characterized by The automatic cultivation execution module comprises: An environment control unit is configured to receive cultivation scheme data and combine them into an environment parameter set, generate control instructions, and output them to temperature and humidity, light, nutrient solution composition and concentration control equipment in the controllable environment cultivation device. A growth monitoring unit is configured to receive raw environment and image data collected by sensors arranged in the controllable environment cultivation device, adopt image recognition and data fusion processing, and output cultivation environment data and plant growth data to the cadmium adsorption evaluation module.

5. The cadmium low-adsorbing potato breeding system according to claim 4, characterized by The growth monitoring unit comprises: An image processing subunit is configured to receive plant raw image data collected by a camera, adopt a deep learning image segmentation algorithm, extract and output plant height and leaf area index growth indicator data to a data fusion subunit. A data fusion subunit is configured to receive plant height and leaf area index growth indicator data, and real-time environment data collected by sensors in the controllable environment cultivation device, perform spatio-temporal alignment and fusion processing, and output time-series aligned cultivation environment data and plant growth data to the cadmium adsorption evaluation module.

6. The cadmium low-adsorbing potato breeding system according to claim 5, characterized by The adsorption capacity evaluation model is: Adsorption capacity score=(1-enrichment coefficient)×α+(1-transport coefficient)×β+biomass coefficient×γ; Wherein α, β, γ are weight coefficients, and α+β+γ=1, the weight can be dynamically adjusted with iterative evaluation results.

7. The cadmium low-adsorbing potato breeding system according to claim 6, characterized by The system further comprises: A user interaction and visualization module is configured to receive user query and instruction data, and receive process data and result data output by each module of the system, perform visualization rendering, and output graphical interface data to a display device.

8. The cadmium low-adsorbing potato breeding system according to claim 7, characterized by The system is deployed on a cloud platform, and the automatic cultivation execution module transmits data with remote cultivation devices and sensor equipment through an Internet of Things protocol, receives control instructions and transmits collected data.

Images