Method for predicting photosynthetic capacity of a plant using leaf traits of the plant

Abstract

The application discloses a method for predicting photosynthetic capacity of plants by using plant leaf traits, and relates to the field of agricultural science and technology, and comprises the following steps: obtaining image data of target plant leaves in a natural state by a high-resolution imaging device, and synchronously recording environmental parameter information at the time of acquisition; extracting multi-dimensional trait characteristic parameters of the leaves based on the image data, including color indexes, texture characteristics, area, thickness estimation values and edge contour morphologies; and using the correlation rules of the environmental parameters and the leaf traits to perform stress characteristic identification on the trait data. The application introduces a stress identification and data correction mechanism, improves the accuracy of photosynthetic capacity prediction, has good environmental adaptability and crop universality, is suitable for various agricultural scenes, constructs a prediction feedback closed-loop system, realizes continuous optimization and intelligent adjustment of the model, and significantly enhances the stability, reliability and practical application value of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural science and technology, and particularly relates to a method for predicting photosynthetic capacity of plants by using leaf traits of plants. BACKGROUND

[0002] The prediction of photosynthetic capacity of plants by using leaf traits of plants refers to the analysis of physical morphology and structural characteristics of plant leaves, such as the size, shape, color, thickness, chlorophyll content, texture structure, etc. of the leaves, to infer or calculate the photosynthetic efficiency of the plant under certain environmental conditions. Because the photosynthetic capacity of plants is directly affected by the structural and functional state of their leaves, these trait parameters can reflect the ability of plants to absorb light energy, exchange gases, and perform photosynthesis. With the aid of image acquisition, sensing technology or AI models, the photosynthetic capacity of plants can be quickly and non-destructively predicted through leaf characteristics without complex experimental determination, thereby improving the monitoring efficiency of crop physiological state and assisting in decision-making in the fields of breeding, precision agriculture and ecological management.

[0003] The prior art has the following disadvantages:

[0004] In the prior art, in the process of predicting photosynthetic capacity by using leaf traits of plants, there is a problem of distortion of trait characteristics due to environmental factors. Such interference is usually caused by high temperature, drought, strong light, pests and diseases or soil salinity, etc. under non-physiological normal state, which can deepen or lighten the color of the leaves, make the texture brittle or curl, and cause abnormal fluctuations in the thickness, etc. These changes are mostly short-term adaptive responses of plants under extreme or stressed environments, and do not necessarily truly reflect the photosynthetic efficiency. Once the prediction system takes these "stress traits" as input without identification and correction, it may incorrectly estimate the photosynthetic capacity level of the plant.

[0005] For example, the system may misjudge the deepening of leaf color as an increase in chlorophyll content, thereby overestimating the photosynthetic intensity. Such misjudgment not only interferes with the accurate assessment of the growth state of crops, but also may cause directional errors in precision agriculture measures, irrigation management or disease warning, ultimately leading to yield reduction, resource waste and even imbalance of farmland ecosystems, etc.

[0006] The above information disclosed in the background section is only used to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0007] The purpose of the present application is to provide a method for predicting the photosynthetic capacity of plants using plant leaf traits, which improves the accuracy of photosynthetic capacity prediction by introducing a stress recognition and data correction mechanism, has good environmental adaptability and crop universality, is suitable for various agricultural scenes, constructs a prediction feedback closed-loop system, realizes continuous optimization and intelligent adjustment of the model, and significantly enhances the stability, reliability and practical application value of the system, to solve the problems in the above background art.

[0008] To achieve the above purpose, the present application provides the following technical scheme: a method for predicting the photosynthetic capacity of plants using plant leaf traits, comprising the following steps:

[0009] An image acquisition step, in which image data of the target plant leaf in a natural state is obtained by a high-resolution imaging device, and environmental parameter information at the time of acquisition is recorded synchronously;

[0010] A leaf trait extraction step, in which multi-dimensional trait feature parameters of the leaf are extracted based on the image data, including color indicators, texture features, area, thickness estimates, and edge contour morphology;

[0011] A stress recognition step, in which stress feature recognition is performed on the trait data using the association rules of environmental parameters and leaf traits, and the leaf samples in an abnormal state are marked by comparing with the preset normal physiological parameter interval;

[0012] A data correction step, in which trait parameter correction is performed on the leaf samples marked as abnormal, the trait deviation value caused by stress reaction is eliminated, and the prediction reference value under the normal physiological state is restored;

[0013] A photosynthetic capacity prediction step, in which the corrected trait parameters are input into the trained prediction model, and the photosynthetic efficiency index of the plant individual under the current physiological state is output;

[0014] A prediction result verification step, in which the deviation rate of the predicted photosynthetic capacity result is evaluated by comparing the actual gas exchange measurement value of the chlorophyll fluorescence parameter of the same plant in the same time period, which is used to feedback and optimize the prediction model accuracy.

[0015] Preferably, the high-resolution imaging device used in the image acquisition step is a multispectral imaging camera, which simultaneously collects visible light, near-infrared, and red edge band image data for capturing the reflection characteristics of the leaf under different wavebands;

[0016] At the same time of collecting images, an integrated environmental perception module is installed to record the light intensity, air temperature and humidity, wind speed, and soil moisture in real time, and match the time stamp with the image file to construct an image and environmental data synchronous acquisition database.

[0017] Preferably, the leaf trait extraction step adopts an image segmentation algorithm to accurately extract the leaf area, and calculates the texture directionality index, gray level co-occurrence matrix energy, contrast, and uniformity index in the leaf area based on convolution filtering and edge detection algorithm;

[0018] In addition, the pixel unit is converted into millimeter unit through image measurement scale calibration technology, so as to accurately estimate the leaf area, perimeter, aspect ratio, and thickness estimate value.

[0019] Preferably, the stress recognition step adopts an environmental dynamic threshold method to recognize the abnormal physiological state, and cross analyzes the measured environmental factor change trend with the historical data by constructing a leaf trait-environmental response dynamic characteristic database, so as to recognize the leaf response mode deviating from the physiological steady state;

[0020] Specifically, under drought conditions, samples with increased leaf gray value and decreased texture directionality are recognized; under high temperature conditions, the abnormal increase of leaf edge curling is monitored; under strong light exposure conditions, the abnormal feature point with sudden increase of color saturation and decrease of average green channel value is recognized.

[0021] Preferably, the data correction step includes stress attribution analysis on abnormal samples, and different correction models are used for parameter adjustment according to different environmental inducements;

[0022] For samples with dark leaf color caused by drought, a reverse fitting method based on the chlorophyll index is used to adjust the color index; for samples with enhanced texture caused by strong light, image blur correction filtering is used to restore the real texture features; for samples with reduced area caused by high temperature curling, morphological reconstruction algorithm is used to estimate the leaf area in the unfolded state.

[0023] Preferably, the photosynthetic capacity prediction step adopts an integrated learning model, which integrates gradient boosting decision tree, support vector regression, and shallow neural network three sub-models, and combines the output prediction results by weighted average method;

[0024] Each sub-model is trained and optimized for different types of leaf trait characteristics. The gradient boosting decision tree is used to model the nonlinear relationship between texture and color features and photosynthetic efficiency, the support vector regression is used to model the boundary distribution of leaf area and thickness and gas exchange capacity, and the shallow neural network is used for fast discriminant prediction after the fusion of all parameter characteristics.

[0025] Preferably, the photosynthetic capacity prediction step introduces a composite feature construction mechanism that integrates leaf image color and morphological parameters, and the specific prediction steps are as follows:

[0026] Based on the gray mean values of red, green, and blue channels extracted from the original image, they are respectively denoted as , and , construct a color parameter fusion value to comprehensively reflect the color expression of the leaf under the light condition, and a calculation formula of the color parameter fusion value is as follows:

[0027]

[0028] , wherein, , and are average pixel gray values in the red channel, the green channel and the blue channel of the plant leaf image respectively, is the color parameter fusion value;

[0029] An area morphological index is calculated, which comprehensively considers the leaf surface area and the edge circumference, and reflects the integrity and expansibility of the leaf morphology through a geometric expression of an ellipse fitting, and a calculation expression is as follows:

[0030]

[0031] , wherein, is the leaf area, is the leaf circumference, is the area morphological index, is the circular ratio;

[0032] A photosynthetic prediction composite feature is constructed through the color parameter fusion value and the area morphological index , which is used as one of the final model input variables, enhances the precise response ability of the model to the leaf health status and the photosynthetic potential, and improves the generalization under the variable environment and the various leaf types, and a construction expression of the photosynthetic prediction composite feature is as follows:

[0033]

[0034] , wherein, is the morphological-color prediction feature combination value.

[0035] Preferably, the fluorescence parameters and the gas exchange parameters synchronously collected are used as reference standards in the prediction result verification step to construct a prediction deviation evaluation index system;

[0036] The prediction deviation evaluation index system includes an error between the predicted value and the measured value, a root mean square error and a model stability fluctuation index;

[0037] When the error between the predicted value and the measured value exceeds a set threshold value, an error tracing mechanism is triggered to retrace each trait parameter and environmental variable in the prediction link, automatically judge the potential reasons for the prediction failure, and generate a model optimization and data cleaning suggestion.

[0038] Preferably, a dynamic error feedback evaluation mechanism is introduced in the prediction result verification step, and the prediction accuracy and model stability are comprehensively considered, and the specific steps are as follows:

[0039] The prediction percentage error is calculated to measure the deviation between the predicted value and the measured value, and the calculation formula is as follows:

[0040]

[0041] In the formula, is the percentage error, is the predicted photosynthetic efficiency, is the measured photosynthetic efficiency;

[0042] The model robustness coefficient is introduced to reflect the output stability of the prediction model under multiple repeated input conditions, and the calculation formula is as follows:

[0043]

[0044] In the formula, is the variance of the predicted output, is the model robustness coefficient;

[0045] The comprehensive percentage error and the model robustness coefficient are used to construct the prediction performance comprehensive evaluation score, which is used to score the entire model prediction system, and the calculation formula is as follows:

[0046]

[0047] In the formula, is the comprehensive evaluation score;

[0048] The comprehensive evaluation score combines accuracy and robustness with equal weight, and when is less than the set threshold value, the backtracking analysis is automatically started, and the trait parameter and environmental factor input source used in this prediction are reexamined to determine whether the model needs to be corrected or the collection strategy needs to be adjusted.

[0049] In the above technical scheme, the technical effects and advantages provided by the present application are as follows:

[0050] The present application significantly improves the prediction accuracy of plant photosynthetic capacity in complex environments by introducing a stress recognition and data correction mechanism. In the prior art, when plants are subjected to drought, high temperature and other stresses, their leaf traits often deviate from the normal physiological state. If not recognized and corrected, it is easy to overestimate or underestimate the photosynthetic capacity. The present application can effectively identify the distortion of traits under stress through dynamic matching analysis of environmental parameters and trait changes, and use the historical physiological model for reasonable correction, ensuring that the trait data entering the prediction model can more truly reflect the photosynthetic physiological level of the plant, thereby reducing the misjudgment rate from the root.

[0051] The present application has good environmental adaptability and multi-crop versatility, and is suitable for deployment and application in various agricultural scenes such as field environment, greenhouse facilities and urban gardening. By integrating image acquisition equipment and multi-channel environmental sensors, a mobile and automatically operated intelligent acquisition platform is constructed, and the system can adapt to plant phenotype acquisition tasks under different light, soil, water and climate conditions. At the same time, the system uses a standardized processing flow for the extraction of leaf traits, which can be quickly migrated to different plant species and strains, has strong versatility and scalability, and provides a solid data foundation for large-scale agricultural informatization, automation and intelligent management.

[0052] The present application not only realizes the prediction from leaf traits to photosynthetic capacity, but also further constructs a prediction result verification and deviation feedback mechanism based on measured data, thereby forming a complete prediction closed-loop system. After the prediction is completed, the system compares with the measured fluorescence parameters or gas exchange data to dynamically evaluate the reliability and stability of the model output; when the prediction deviation exceeds the threshold, the system can automatically locate the source of the prediction error, adjust the model parameters or input features, and improve the adaptability and self-evolution ability of the model. This closed-loop intelligent mechanism enables the entire system to have continuous learning and optimization capabilities, and can maintain high precision operation in long-term application, greatly improving its practical value in precision agriculture and plant phenotype research. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0054] Figure 1 The present application is a method for predicting plant photosynthetic capacity using plant leaf traits. DETAILED DESCRIPTION

[0055] Example implementations are now described with reference to the drawings. Example implementations can, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive gist to those skilled in the art.

[0056] The present application provides a method for predicting photosynthetic capacity of plants using leaf traits of plants, as shown in the following steps: Figure 1 The present application provides a method for predicting photosynthetic capacity of plants using leaf traits of plants, as shown in the following steps:

[0057] An image acquisition step, in which image data of the target plant leaf in a natural state is acquired by a high-resolution imaging device, and environmental parameter information at the time of acquisition is recorded synchronously;

[0058] The high-resolution imaging device used in the image acquisition step is a multispectral imaging camera, which simultaneously acquires visible light, near-infrared, and red edge band image data for capturing the reflection characteristics of the leaf under different wavebands.

[0059] At the same time of acquiring the image, an integrated environment perception module is installed to record the light intensity (unit: μmol·m-2·s-1), air temperature and humidity (unit: ℃ and %), wind speed (unit: m / s), and soil moisture (unit: %) in real time, and match the time stamp with the image file to construct an image and environment data synchronous acquisition database.

[0060] Through this means, the synchronization and accuracy of data fusion are improved, and time-effective and consistent environmental basic data support is provided for subsequent trait identification and stress judgment.

[0061] A leaf trait extraction step, in which multi-dimensional trait characteristic parameters of the leaf are extracted based on the image data, including color indicators, texture characteristics, area, thickness estimates, and edge contour morphology.

[0062] In the leaf trait extraction step, an image segmentation algorithm is used to accurately extract the leaf area, and based on convolution filtering and edge detection algorithms, texture directionality indicators, gray level co-occurrence matrix energy, contrast, and uniformity indicators in the leaf area are calculated.

[0063] In addition, the pixel unit is converted to millimeter unit through image measurement scale calibration technology, so as to accurately estimate the leaf area, perimeter, aspect ratio, and thickness estimate.

[0064] The above trait parameters are subjected to normalization processing to maintain numerical distribution consistency before model input, improving the model recognition ability and stability. This step also includes automatic correction of the imaging light angle to reduce shadow interference caused by side light and improve the consistency and reliability of leaf trait extraction.

[0065] ​The stress recognition step uses the association rules between environmental parameters and leaf traits to identify the stress characteristics of the trait data, and marks the leaf samples in an abnormal state by comparing with the preset normal physiological parameter interval;

[0066] The environmental dynamic threshold method is used in the stress recognition step to identify the abnormal physiological state, the cross analysis of the measured environmental factor change trend and the historical data is performed through the construction of the leaf trait-environment response dynamic characteristic database, and the leaf response mode deviating from the physiological steady state is identified;

[0067] Specifically, under drought conditions, samples with increased leaf gray value and decreased texture directionality are identified; under high temperature conditions, the abnormal increase of leaf edge curling is monitored; under strong light exposure conditions, the abnormal feature points of sudden increase of color saturation and decrease of average green channel value are identified.

[0068] Through the multi-dimensional index combination verification mode, the stress-induced distortion samples are calibrated, and the stress interference recognition capability is enhanced.

[0069] The data correction step corrects the trait parameters of the leaf samples marked as abnormal, eliminates the trait deviation value caused by stress response, and restores the prediction reference value under normal physiological state;

[0070] The stress attribution analysis of abnormal samples is included in the data correction step, and different correction models are used for parameter adjustment according to different environmental inducements;

[0071] For samples with dark leaf color caused by drought, the color index is adjusted by using the reverse fitting based on the chlorophyll index; for samples with enhanced texture caused by strong light, the real texture features are restored by using image blur correction filtering; for samples with area reduction caused by high temperature curling, the leaf area in the unfolded state is estimated using morphological reconstruction algorithm.

[0072] Each type of correction operation is based on the modeling of the same type of healthy samples to form a prediction reference value, ensuring that the correction value is returned to a reasonable range rather than simply normalized, improving the physiological reasonableness and prediction effectiveness of the correction.

[0073] The photosynthetic capacity prediction step inputs the corrected trait parameters into the trained prediction model, and outputs the photosynthetic efficiency index of the plant individual under the current physiological state;

[0074] The integrated learning model is used in the photosynthetic capacity prediction step, which combines gradient boosting decision tree (GBDT), support vector regression (SVR) and shallow neural network (SNN) three sub-models, and outputs the prediction results by weighted average combination;

[0075] Each sub-model is trained and optimized for different types of leaf trait characteristics. Gradient boosting decision tree (GBDT) is used to model the nonlinear relationship between texture and color features and photosynthetic efficiency. Support vector regression (SVR) is used to model the boundary distribution of leaf area and thickness and gas exchange capacity. Shallow neural network (SNN) is used for fast discriminant prediction after feature fusion.

[0076] The combination weight is dynamically optimized by cross-validation method to ensure the stability and high accuracy of prediction performance in a variable environment and diverse leaf morphology.

[0077] In the photosynthetic capacity prediction step, a composite feature construction mechanism is introduced to fuse leaf image color and morphological parameters. The specific prediction steps are as follows:

[0078] Based on the gray mean values of red, green and blue channels extracted from the original image, denoted as , and , a color parameter fusion value is constructed to comprehensively reflect the color expression of the leaf under the light condition. The calculation formula of the color parameter fusion value is as follows:

[0079]

[0080] , where , and are the average pixel gray values of the red channel (Red), green channel (Green) and blue channel (Blue) in the plant leaf image, respectively, with a value range of 0-255. After image segmentation of the leaf area, the average gray values of all pixels in each channel are calculated, The green channel is the most representative in plant photosynthetic capacity analysis, as chlorophyll mainly reflects green light and is an important basis for judging the physiological state of plants, and Supplementary spectral reflectance characteristics help to determine whether there are color changes such as disease and aging. The green channel coefficient in the weight is 2, highlighting its sensitivity to chlorophyll content in plant photosynthesis, is the color parameter fusion value, which is the weighted average result of the gray values of the red, green and blue channels, used to represent the average response intensity of the leaf color under different spectra. The higher the value, the stronger the leaf reflectivity (possibly related to high chlorophyll content and healthy surface); abnormal values may indicate stress response (such as yellowing and loss of green);

[0081] The weighted green channel is used twice to reflect its dominant role in plant chlorophyll response and enhance the representativeness of the formula for leaf photosynthetic potential.

[0082] The area morphology index is calculated, which comprehensively considers the leaf surface area and its edge perimeter. It reflects the integrity and extensibility of the leaf morphology by fitting an approximate ellipse geometric expression. The calculation expression is as follows:

[0083]

[0084] In the formula, Leaf area represents the actual projected area of ​​a target plant's leaf in an image. It is calculated after image pixel-to-physical unit conversion. After image segmentation, the total number of leaf pixels is calculated, multiplied by the area per unit pixel, and converted to the actual area. It is a basic structural factor characterizing the photosynthetic potential of a plant. The larger the leaf, the greater the potential photosynthetic surface area, and generally the stronger its photosynthetic capacity. The blade perimeter represents the length of the closed contour of the blade edge. After obtaining the set of blade edge points based on an image edge detection algorithm, the length of the contour curve is calculated, characterizing the complexity of the blade edge morphology. Stress conditions such as curling or shrinkage can cause changes in the perimeter, affecting the prediction model's judgment. It is an area morphology index, a geometric structure index constructed by combining leaf area and perimeter. It integrates area and perimeter into a morphological feature to approximately reflect the "degree of unfolding" and integrity of the leaf. The larger the value, the fuller and healthier the leaf shape usually is. Abnormal values ​​can reflect structural changes such as curling and disease. Pi is the ratio of a circle's circumference to its diameter, and its approximate value is 3.14159. The purpose of its introduction is to increase the area of ​​the leaf. With perimeter The combination of these factors is standardized into a morphological index that compares to an approximately circular structure; this structure is similar to the calculation of the shape factor, which is widely used in image analysis to measure the degree to which a region deviates from a standard circle. This is achieved by dividing by... This can more accurately reflect whether the leaves are close to a geometrically regular state, thereby improving the discriminative power of morphological information in predicting photosynthetic capacity;

[0085] This formula helps to establish a unified structural measurement benchmark across different varieties and growth stages, suppressing the interference of shape differences on the model.

[0086] By color parameter blending value and area morphology index A composite feature for photosynthesis prediction is constructed and used as one of the input variables of the final model, as shown in the following formula:

[0087]

[0088] In the formula, is a morphological-color predictive feature combination value, and the color parameter fusion value and an area morphological index After multiplication and natural logarithm taking, a composite predictive feature variable is formed, which is one of the core features of the model input, helps to enhance the precise response ability of the model to the leaf health status and photosynthetic potential, and significantly improves the generalization under variable environment and various leaf types.

[0089] Through the above steps, the identification and modeling ability of the model for the nonlinear relationship between complex leaf morphology and optical characteristics is significantly improved.

[0090] The prediction result verification step compares the chlorophyll fluorescence parameters and gas exchange measured values of the same plant in the same time period to evaluate the deviation rate of the predicted photosynthetic capacity result, which is used to feedback and optimize the prediction model accuracy;

[0091] In the prediction result verification step, the simultaneously collected fluorescence parameters (such as Fv / Fm value, photochemical quenching qP, and non-photochemical quenching NPQ) and gas exchange parameters (such as net photosynthetic rate Pn and stomatal conductance Gs) are used as reference standards to construct a prediction deviation evaluation index system;

[0092] The prediction deviation evaluation index system includes the error between the predicted value and the measured value, the root mean square error, and the model stability fluctuation index (coefficient of variation);

[0093] When the error between the predicted value and the measured value exceeds the set threshold, an error tracing mechanism is triggered to retrace each trait parameter and environmental variable in the prediction link, automatically determine the potential cause of prediction failure, and generate model optimization and data cleaning suggestions.

[0094] In the prediction result verification step, a dynamic error feedback evaluation mechanism is introduced to comprehensively consider the prediction accuracy and model stability two key dimensions, and the specific steps are as follows:

[0095] Calculate the prediction percentage error to measure the deviation degree between the predicted value and the measured value, and the calculation formula is as follows:

[0096]

[0097] In the formula, is the percentage error, which represents the deviation degree between the predicted value and the true value, and is expressed in percentage. The absolute value is used to ensure that both overestimation and underestimation can be identified, and the smaller the value, the higher the prediction accuracy, is the predicted photosynthetic efficiency, which represents the rate of fixing carbon dioxide per unit area per unit time, and the unit is , is the measured photosynthetic efficiency, which is the real photosynthetic capacity data obtained by professional instruments (such as LI-6400 photosynthetic determination system) for field gas exchange determination of target plants, and the unit is ;

[0098] This error reflects the deviation of the prediction system at the current sample point, and is a basic index for evaluating the prediction accuracy.

[0099] The model robustness coefficient is introduced to reflect the output stability of the prediction model under multiple repeated inputs, and the calculation formula is as follows:

[0100]

[0101] , wherein, is the variance of the prediction output, which refers to the numerical fluctuation degree between multiple prediction results output by the model when the same model input (or repeated input under slight disturbance) is used for a group of same type plant samples, and the unit is the square of the photosynthetic efficiency, which is used to measure the consistency of the model output, is the model robustness coefficient, which is a quantitative reflection of the prediction stability, and the value closer to 1 indicates that the model output is more stable, that is, it is not sensitive to slight changes in input, which means that the model has good anti-interference ability, and the smaller the value, the greater the fluctuation in multiple predictions, and the poorer the stability;

[0102] The comprehensive percentage error and the model robustness coefficient are used to construct the prediction performance comprehensive evaluation score, which is used to score the whole model prediction system, and the calculation formula is as follows:

[0103]

[0104] , wherein, is the comprehensive evaluation score, which is the weighted average result of the prediction accuracy and the robustness of the model according to equal weight. The higher the value, the better the overall performance of the model in the current task, and if is lower than the set threshold, it means that the model performs poorly, and error backtracking analysis and model optimization are needed;

[0105] The comprehensive evaluation score combines accuracy and robustness with equal weight, and when is less than the set threshold, the backtracking analysis is automatically started, and the input source of the trait parameters and environmental factors used for this prediction is reexamined to determine whether the model needs to be corrected or the collection strategy needs to be adjusted.

[0106] This mechanism ensures that the prediction system has self-correction ability in long-term operation, improves the reliability and adaptability of use.

[0107] Below is a detailed expansion of the above three specific embodiments, each part exceeds 1000 words, fully describes the technical process and implementation details from device building to algorithm processing, system linkage and field deployment:

[0108] Embodiment one: This embodiment mainly faces the demand for efficient and standardized collection and analysis of plant population leaf traits in large-scale agricultural planting scenarios. It uses unmanned collection platform combined with multispectral imaging and high-precision environmental monitoring technology to realize rapid extraction and data preprocessing of leaf traits.

[0109] Firstly, an integrated field monitoring platform is established. This platform uses an automatic driving agricultural vehicle or a drone as a carrier, carries a multispectral camera (including visible light, near-infrared, red edge band), which has a resolution of at least 300 dpi and has automatic focusing and exposure control functions, which can meet the image collection needs under different light conditions. To improve the timeliness and continuous operation ability of the system, the platform is equipped with a high-performance edge computing module for real-time image preprocessing to reduce data transmission delay.

[0110] The collection platform synchronously integrates environmental perception modules, including air temperature and humidity sensors, solar radiation intensity meters, anemometers, soil temperature and humidity sensors, etc. All sensors and image acquisition systems share timestamps, and are bound through a unified data bus and database to ensure that each set of leaf image data can correspond to complete environmental condition records. This ensures data integrity and provides a solid foundation for subsequent environmental factor-driven analysis.

[0111] During data collection, the platform moves along the preset trajectory and collects plants point by point in the specified row and column area. The system identifies the plant area through image automatic calibration, distinguishes the target leaf from non-leaf areas (such as background soil and crop stems) in each image, and extracts the main body of the leaf using image segmentation algorithms (such as U-Net, DeepLab). After extracting the leaf area, the system performs standardized analysis and calculates multiple trait parameters, including:

[0112] Color features: extract RGB channel mean, color saturation, and brightness (HSV model) to estimate chlorophyll content;

[0113] Texture features: use gray level co-occurrence matrix method to calculate energy, contrast, uniformity, and entropy, etc. to reflect leaf tissue state;

[0114] Geometric features: including leaf area, leaf length-width ratio, edge serration rate, curling coefficient, thickness estimation, etc. to evaluate leaf growth integrity;

[0115] Waveband reflectance indices: such as NDVI (Normalized Difference Vegetation Index), GNDVI, RENDVI, etc., used to estimate photosynthetic efficiency indices;

[0116] The system unifies and normalizes the above multi-dimensional features, and constructs a complete database structure of "leaf number-image file- trait parameter-environment variable". The data is backed up to the local server and the cloud platform at the same time, supporting subsequent batch processing, analysis modeling and model training.

[0117] This embodiment is particularly suitable for crop variety comparison test, breeding project phenotype data accumulation, agricultural digital twin modeling and other scenes, and can quickly build standardized and high-precision crop trait portraits to provide key support for precision agriculture.

[0118] Embodiment two: In actual agricultural production, plants often face various sudden or continuous non-physiological environmental stresses, such as extreme temperature, drought, salt and alkali stress, diseases and pests, etc. These stress factors often cause atypical changes in plant leaf traits, such as color deepening, edge drying, abnormal thickness fluctuation, etc. If such trait parameters are directly used for photosynthetic capacity prediction, it is easy to cause the result to be distorted. Therefore, this embodiment proposes a stress recognition and correction mechanism based on dual judgment of environment and traits, to improve the robustness of the prediction system.

[0119] The system first constructs a "stress condition discrimination threshold table" from environmental data, combines agricultural ecological zoning standards and crop growth physiological thresholds, and defines multiple stress discrimination rules. For example: the daily average temperature of the location where the crop is located exceeds 35°C for more than 48 hours; the number of consecutive days without effective precipitation is more than 7 days and the soil moisture content is less than 20%; the solar radiation intensity exceeds 1500 The system matches the environmental data at the time of image acquisition with the above rules to determine whether the collected sample is likely to be in a stressed state.

[0120] On this basis, the system further identifies stress traits in combination with trait data. Taking drought stress as an example, the system identifies samples with significantly increased gray value and significantly decreased green channel value by comparing the leaf color change trajectory with the typical performance database of water stress; taking heat stress as an example, the system identifies samples with increased edge curvature coefficient and reduced leaf area. The system uses a multi-index comprehensive scoring method to score the "stress probability" of the sample, and sets a marking threshold. When the score is higher than the critical value, the sample will be marked with the "need to correct" label.

[0121] After entering the correction stage, the system introduces a historical healthy sample comparison method, selects a "healthy sample group" with similar varieties, growth periods, and soil conditions but in a non-stressed environment, and performs regression estimation correction on the current sample traits. For example, the green channel value is adjusted according to the historical function relationship between the green channel value and NDVI using linear regression; the area reduction caused by curled leaves under heat stress is inversely calculated using a leaf simulation unfolding model to compensate for the proper area value.

[0122] All corrected data is finally summarized to generate a "corrected trait data set", and the version information and calculation process of the original collected data and the corrected data are retained for subsequent model training difference analysis. This mechanism can effectively eliminate "false traits" caused by environmental stress, ensure that the trait parameters input into the model more truly reflect the physiological basis state of the crop, and improve the stability and adaptability of the prediction system in complex field environments.

[0123] Embodiment three: This embodiment constructs a self-learning type prediction and verification closed-loop system, uses multi-model fusion technology to realize high-precision prediction of plant photosynthetic capacity, and continuously optimizes system performance through error tracking and feedback mechanism, which is particularly suitable for long-term monitoring, fine management and production prediction needs.

[0124] First, the system receives the leaf trait parameter data after stress recognition and correction, and constructs an input vector according to the model type. In order to ensure the generalization ability of the model, the system uses an ensemble learning architecture to integrate three sub-models: gradient boosting decision tree (GBDT), support vector regression (SVR), and shallow neural network (SNN). Among them, GBDT is good at handling nonlinear relationships and feature importance sorting, SVR is suitable for capturing boundary outliers, and SNN is responsible for modeling the nonlinear cross-interaction between parameters. After the three models are trained and independent prediction results are generated, the system performs weighted fusion according to the set weight to obtain the final photosynthetic efficiency prediction value (unit: μmol / m2s). ).

[0125] In order to ensure the credibility and traceability of the prediction results, the system synchronously configures a physiological measurement module to randomly select part of the samples for fluorescence parameter measurement (such as maximum photochemical efficiency Fv / Fm, qP, NPQ) and gas exchange instrument measurement (such as net photosynthetic rate Pn, transpiration rate Tr). After uploading the measurement results, the system automatically compares the predicted value with the measured value, calculates the key error indicators: absolute error, relative error RE, root mean square error RMSE, model output variance (stability index), etc.

[0126] If the system detects that the relative error of multiple consecutive samples exceeds 10%, the error tracking mechanism is started. The system automatically tracks the correlation between each trait parameter and the final prediction bias, and identifies key feature variables that may cause errors, such as abnormal color channel values, estimated curling rate bias, etc., prompting the user to perform model retraining, logic update or parameter importance weight adjustment.

[0127] The system also has a "continuous learning" capability, and the user can choose to add each round of validation data as a feedback sample to the model incremental learning pool, and perform local weight adjustment and re-fitting under the original model structure, thereby realizing the self-adaptation of the prediction model to new environments and crop states.

[0128] This linkage prediction-verification mechanism forms a closed loop of data flow, model flow, and feedback flow, which not only improves the prediction accuracy, but also significantly improves the model stability and agricultural scene practicability, providing strong support for intelligent decision-making in agricultural production.

[0129] The present application significantly improves the prediction accuracy of plant photosynthetic capacity in complex environments by introducing stress recognition and data correction mechanisms. In the prior art, when plants are under stress such as drought and high temperature, their leaf traits often deviate from normal physiological state. If not identified and corrected, it is easy to overestimate or underestimate the photosynthetic capacity. The present application can effectively identify the distortion of traits under stress through dynamic matching analysis of environmental parameters and trait changes, and use historical physiological models for reasonable correction, ensuring that the trait data entering the prediction model can more truly reflect the photosynthetic physiological level of the plant, thereby reducing the misjudgment rate from the root.

[0130] The present application has good environmental adaptability and multi-crop versatility, and is suitable for deployment and application in various agricultural scenarios such as field environment, greenhouse facilities, and urban gardening. By integrating image acquisition equipment and multi-channel environmental sensors, an intelligent acquisition platform that can be moved and automatically operated is constructed, and the system can adapt to plant phenotype acquisition tasks under different light, soil, water, and climate conditions. At the same time, the system uses a standardized processing flow for leaf trait extraction, which can be quickly migrated to different plant species and strains, has strong versatility and scalability, and provides a solid data foundation for large-scale agricultural informatization, automation, and intelligent management.

[0131] The present application not only realizes the prediction from leaf traits to photosynthetic capacity, but also further constructs a prediction result verification and deviation feedback mechanism based on measured data, thereby forming a complete prediction closed-loop system. After the prediction ends, the system compares with the measured fluorescence parameters or gas exchange data to dynamically evaluate the reliability and stability of the model output; when the prediction deviation exceeds the threshold, the system can automatically locate the source of the prediction error, adjust the model parameters or input features, and improve the adaptability and self-evolution ability of the model. This closed-loop intelligent mechanism enables the entire system to have continuous learning and optimization capabilities, can maintain high precision operation in long-term application, and greatly improves its practical value in precision agriculture and plant phenotype research.

[0132] The above formulas are dimensionless numerical calculations, the formulas are obtained by collecting a large amount of data to simulate a formula of the latest real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0133] The above only describes some exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present application. Therefore, the above figures and descriptions are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

[0134] It should be noted that in this paper, if there are relationship terms such as first and second, they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the sentence "includes a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0135] It should be understood that in various embodiments of the present application, the size of the sequence number of each process described above does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0136] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0137] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0138] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, i.e. they can be located in one place or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment according to actual needs.

[0139] In addition, the functional units in each embodiment of the present application can be integrated into one processing unit, or each unit can be physically present separately, or two or more units can be integrated into one unit.

[0140] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any skilled person in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0141] The above only describes some exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without deviating from the spirit and scope of the present application. Therefore, the above figures and descriptions are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

Claims

1. A method for predicting photosynthetic capacity of a plant using leaf traits of the plant, characterized in that, The method comprises the following steps: An image acquisition step, in which image data of a target plant leaf in a natural state is acquired by a high-resolution imaging device, and environmental parameter information at the time of acquisition is recorded synchronously; A leaf trait extraction step, in which multi-dimensional trait characteristic parameters of the leaf are extracted based on the image data, including color indicators, texture characteristics, area, thickness estimation values and edge contour morphology; A stress recognition step, in which stress characteristic recognition is performed on the trait data by using a correlation rule of the environmental parameters and the leaf traits, and by comparing with a preset normal physiological parameter interval, leaf samples in an abnormal state are marked; A data correction step, in which trait parameter correction is performed on the leaf samples marked as abnormal, trait deviation values caused by stress response are eliminated, and prediction reference values in a normal physiological state are restored; A photosynthetic capacity prediction step, in which the corrected trait parameters are input into a trained prediction model, and an photosynthetic efficiency indicator of a plant individual in a current physiological state is output; In the photosynthetic capacity prediction step, a composite feature construction mechanism fusing leaf image color and morphological parameters is introduced, and the specific prediction steps are as follows: Based on the gray mean values of red, green and blue channels extracted from the original image, respectively denoted as , and , a color parameter fusion value is constructed to comprehensively reflect the color expression of the leaf under the light condition, and the calculation formula of the color parameter fusion value is as follows: , wherein, , and are the average pixel gray values of the red channel, the green channel and the blue channel in the plant leaf image, respectively, is the color parameter fusion value; An area morphological index is calculated, the area morphological index comprehensively considers the leaf surface area and the edge circumference, and the integrity and expansibility of the leaf morphology are reflected by fitting the geometric expression of an ellipse, and the calculation expression is as follows: wherein is the leaf area, is the leaf perimeter, is the area form index, is the pi; By the color parameter fusion value And the area shape index The photosynthetic prediction composite feature is constructed as one of the final model input variables, which enhances the precise response ability of the model to the leaf health state and its photosynthetic potential, and improves the generalization under the variable environment and various leaf types. The construction expression of the photosynthetic prediction composite feature is as follows: , wherein is a modality-color prediction feature combination value; A prediction result verification step, in which by comparing the actual gas exchange measured values of the chlorophyll fluorescence parameters of the same plant in the same time period, the deviation rate of the predicted photosynthetic capacity result is evaluated, and is used for feedback optimization of the prediction model precision; In the prediction result verification step, a dynamic error feedback evaluation mechanism is introduced, two key dimensions of prediction accuracy and model stability are comprehensively considered, and the specific steps are as follows: A prediction percentage error is calculated, which is used to measure the deviation degree between the predicted value and the measured value, and the calculation formula is as follows: wherein, is the percentage error, is the predicted photosynthetic efficiency, is the measured photosynthetic efficiency; A model robustness coefficient is introduced, which is used to reflect the output stability of the prediction model under multiple repeated input conditions, and the calculation formula is as follows: wherein, is the variance of the prediction output, is a model robustness coefficient; The comprehensive percentage error And the model robustness coefficient The prediction performance comprehensive evaluation score is constructed for the unified scoring of the whole model prediction system, and the calculation formula is as follows: , wherein, is the overall evaluation score; The comprehensive evaluation score combines accuracy and robustness with equal weight, when When less than a set threshold, automatically start backtracking analysis, re-examine the trait parameters and environmental factors input source used in this prediction, to determine whether the model needs to be corrected or the collection strategy needs to be adjusted.

2. The method for predicting photosynthetic capacity of a plant using leaf traits of the plant according to claim 1, characterized in that, The high-resolution imaging device used in the image acquisition step is a multispectral imaging camera, which simultaneously collects visible light, near-infrared and red edge band image data, and is used to capture the reflection characteristics of the leaf under different wavebands; At the same time of collecting images, an integrated environment perception module is installed, which is used to record the light intensity, air temperature and humidity, wind speed and soil moisture in real time, and match the time stamp with the image file, and construct an image and environment data synchronous acquisition database.

3. The method of predicting photosynthetic capacity of a plant using leaf traits of the plant according to claim 1, wherein, In the leaf trait extraction step, an image segmentation algorithm is used to accurately extract the leaf area, and based on convolution filtering and edge detection algorithm, texture directionality indicators, gray level co-occurrence matrix energy, contrast and uniformity indicators in the leaf area are calculated; Through image measurement scale calibration technology, pixel units are converted into millimeter units, and leaf area, circumference, aspect ratio and thickness estimation values are accurately estimated.

4. The method of predicting photosynthetic capacity of a plant using leaf traits of the plant according to claim 1, wherein, In the stress recognition step, an environmental dynamic threshold method is used to identify non-normal physiological state, by constructing a leaf trait-environment response dynamic feature database, the measured environmental factor change trend is cross-analyzed with historical data, and the leaf response mode deviating from the physiological steady state is identified; Specifically, it includes: Under drought conditions, samples with increased leaf gray value and decreased texture directionality are identified. Monitor the abnormal increase of leaf edge curling under high temperature conditions; Identify the abnormal feature points of sudden increase in color saturation and decrease in average green channel value under strong light exposure conditions.

5. The method of predicting photosynthetic capacity of a plant using leaf traits of the plant according to claim 1, wherein, The data correction step includes: Stress attribution analysis is performed on abnormal samples, and different correction models are used for parameter adjustment according to different environmental inducements; For samples with dark leaf color caused by drought, adjust the color index based on the reverse fitting of the chlorophyll index; For samples with enhanced texture caused by strong light, use image blur correction filtering to restore the true texture features; For samples with reduced area caused by high temperature curling, use morphological reconstruction algorithm to estimate the area of the unfolded leaf.

6. The method of predicting photosynthetic capacity of a plant using leaf traits of the plant according to claim 1, wherein, The photosynthetic capacity prediction step uses an integrated learning model that combines gradient boosting decision tree, support vector regression, and shallow neural network sub-models to output prediction results through weighted averaging. Each sub-model is trained and optimized for different types of leaf characteristics. Gradient boosting decision tree is used to model the nonlinear relationship between texture and color features and photosynthetic efficiency. Support vector regression is used to model the boundary distribution of leaf area and thickness and gas exchange capacity. Shallow neural network is used for fast prediction after fusion of all parameters.

7. The method of predicting photosynthetic capacity of a plant using leaf traits of the plant according to claim 1, wherein, In the prediction result verification step, the simultaneously collected fluorescence parameters and gas exchange parameters are used as reference standards to construct a prediction bias evaluation index system. The prediction bias evaluation index system includes the error between the predicted value and the measured value, the root mean square error, and the model stability fluctuation index. When the error between the predicted value and the measured value exceeds the set threshold, the error tracing mechanism is triggered, and each characteristic parameter and environmental variable in the prediction chain is retraced to automatically determine the potential causes of prediction failure for model optimization and data cleaning suggestion generation.

Images