Fruit tree canopy phenotype monitoring method and system based on multispectral image

Abstract

The invention belongs to the technical field of agricultural information, and discloses a fruit tree canopy phenotype monitoring method and system based on a multispectral image, and the method comprises the steps: collecting the original image data of a farmland in an arid region through a multispectral sensor, and processing the spectral channel difference through a threshold segmentation algorithm to preliminarily distinguish a fruit tree region from a soil background; obtaining a preliminary separation result; based on the preliminary separation result, obtaining an optimized fruit tree region; extracting a leaf pixel ratio from the optimized fruit tree region, and judging leaf coverage by calculating a pixel number ratio to obtain a leaf coverage index; determining a biomass distribution diagram based on a leaf coverage index; integrating multi-temporal image data through a biomass distribution diagram, tracking a change trend by adopting time sequence analysis, judging dynamic changes of growth indexes, and obtaining fruit tree growth state evaluation; and for fruit tree growth state evaluation, fusing the leaf coverage index and the biomass distribution diagram to generate a comprehensive phenotypic characteristic report, and determining a final monitoring result.

Description

Technical Field

[0001] This invention belongs to the field of agricultural information technology, specifically relating to a method and system for monitoring fruit tree canopy phenotypic patterns based on multispectral images. Background Technology

[0002] In agricultural production, monitoring the growth status of fruit trees is crucial for ensuring fruit safety and increasing yield. This is especially true in arid regions where the growing environment for fruit trees is more complex, making timely understanding of their health status and growth characteristics particularly urgent. Studying fruit tree phenotypic characteristics, such as leaf cover and overall biomass, not only reflects the fruit trees' adaptability to the environment but also provides important data for precision agricultural management. However, research and application in this field still face many challenges and urgently require breakthroughs in existing technological limitations.

[0003] Currently, methods for monitoring fruit tree growth largely rely on ground measurements or remote sensing alone. These methods often fall short of requirements in terms of coverage and real-time performance. This is especially true in arid regions with complex terrain and uneven fruit tree distribution, where traditional methods struggle to comprehensively capture growth changes across large areas. Furthermore, existing technologies are inadequate in handling environmental disturbances, such as variations in light intensity or soil background, frequently leading to inaccurate monitoring results. These limitations prevent agricultural managers from obtaining reliable information in a timely manner, thus impacting decision-making efficiency.

[0004] From a technical perspective, a core challenge in monitoring fruit trees in arid regions is accurately extracting their phenotypic characteristics in complex environments. Particularly when acquiring information through imaging technology, environmental noise becomes a primary problem. For example, the interference between soil reflected light and the spectral characteristics of fruit tree leaves makes accurately distinguishing fruit trees from the background extremely difficult. Furthermore, this interference directly affects the calculation of fruit tree growth indicators; for instance, estimates of leaf cover are often distorted by background noise, resulting in final data that fails to accurately reflect the state of the fruit trees.

[0005] Therefore, accurately separating the image features of fruit trees from the background under complex environmental interference, and precisely calculating growth indicators based on this, has become a key issue in fruit tree monitoring in arid regions. Solving this problem will directly affect the scientific nature and effectiveness of agricultural management, providing crucial support for improving the adaptability of agricultural production in arid areas. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a method and system for monitoring fruit tree canopy phenotypic patterns based on multispectral images. It integrates multi-dimensional data analysis methods to achieve precise monitoring throughout the entire process from image processing to dynamic evaluation, significantly improving the accuracy and reliability of fruit tree growth status assessment in arid regions and providing a scientific basis for agricultural management.

[0007] To achieve the above objectives, the present invention provides the following solution: A method for monitoring fruit tree canopy phenotypic patterns based on multispectral images, the method comprising: Raw image data of farmland in arid areas were collected by multispectral sensors. Threshold segmentation algorithm was used to process spectral channel differences to initially distinguish fruit tree areas from soil background and obtain preliminary separation results. Based on the preliminary separation results, the optimized fruit tree region was obtained; The leaf pixel ratio is extracted from the optimized fruit tree area, and the leaf coverage is determined by calculating the pixel number ratio to obtain the leaf coverage index. Based on leaf cover index, determine biomass distribution map; By integrating multi-temporal image data through biomass distribution maps and using time-series analysis to track change trends, determine the dynamic changes in growth indicators, and obtain an assessment of fruit tree growth status; For the assessment of fruit tree growth status, a comprehensive phenotypic report is generated by integrating leaf cover index and biomass distribution map to determine the final monitoring results.

[0008] Preferably, the method for acquiring raw image data of farmland in arid areas using a multispectral sensor, and then using a threshold segmentation algorithm to process spectral channel differences to initially distinguish fruit tree areas from the soil background, and obtaining preliminary separation results includes: Temporal and spectral feature analysis was performed on farmland image data to determine the optimal classification features, which included: Normalized Difference Vegetation Index (NDVI), Normalized Difference Green-Blue Index (NGBDI), Modified Ratio Vegetation Index (MSR), and Red Edge Band Reflectance. A decision tree classifier was constructed based on the Normalized Difference Vegetation Index (NDVI), the Normalized Green-Blue Difference Index (NGBDI), the Modified Ratio Vegetation Index (MSR), and the red-edge band reflectance. The optimal threshold for each spectral channel was determined through repeated experiments. Based on the optimal threshold and the decision tree classifier, the separation of fruit trees from soil and weeds was achieved.

[0009] Preferred methods for obtaining optimized fruit tree regions based on preliminary separation results include: Based on the preliminary separation results, the spectral reflectance distribution of fruit tree pixels and background pixels is obtained. If the reflectance value of the fruit tree pixels is higher than the preset threshold, it is marked as an effective tree area; otherwise, it is classified as a background area to obtain a refined fruit tree mask. Support vector machine is used to classify edge pixels in the refined fruit tree mask. For confusing pixels caused by environmental interference, the classification boundary is determined by training spectral feature vectors to obtain the optimized fruit tree region.

[0010] Preferably, the method for extracting the leaf pixel ratio from the optimized fruit tree area, determining the leaf coverage by calculating the pixel count ratio, and obtaining the leaf coverage index includes: By collecting image data of the fruit tree area, raw image data is obtained, and the raw image data is preprocessed to obtain the image content after preliminary cleaning. Based on the image content after preliminary cleaning, the leaf pixel region is separated using segmentation technology, and the pixel data related to the leaf is extracted to determine the distribution range of the leaf pixels. If the extracted leaf pixel distribution range meets the preset clarity standard, then the ratio of the number of leaf pixels to the total number of pixels in the fruit tree area is further calculated to obtain the leaf pixel ratio value. By analyzing the pixel ratio values ​​of the leaves and comparing them with preset ratio thresholds, the level of leaf coverage is determined. If the leaf coverage level is lower than the preset minimum threshold, the image data will be processed again to obtain more refined leaf pixel data and determine the correction value of the coverage. Based on the revised coverage value and the grading standards, the final leaf coverage index is obtained.

[0011] Preferred methods for determining biomass distribution maps based on leaf cover indices include: Based on the leaf coverage index, the average spectral intensity value of the fruit tree area is obtained. If the average intensity value exceeds the preset threshold, it is considered a high biomass area; otherwise, it is considered a low biomass area, and a biomass distribution map is determined.

[0012] Preferably, the method for assessing the growth status of fruit trees by integrating multi-temporal image data through biomass distribution maps, using time-series analysis to track trends, and judging the dynamic changes of growth indicators includes: For the biomass distribution map, time series analysis was applied to process multi-temporal data, track the changing trends at each time point, and obtain serialized data of dynamic index values. If the serialized data of the dynamic indicator values ​​fluctuates within the preset threshold range, it is judged that the growth state is stable. If the value exceeds the preset threshold range, it is marked as an abnormal state, and a preliminary growth status assessment result is obtained. Based on the preliminary growth status assessment results and combined with the background information of crop growth status, the characteristics of key time node index changes are extracted to determine the specific classification of growth stages. By classifying the specific growth stages, the support vector machine model is used to further analyze the dynamic index values, determine the health status of crop growth, and obtain a comprehensive judgment result. After obtaining the comprehensive judgment results, for areas with abnormal conditions, the corresponding image data sources and biomass distribution information are extracted to generate targeted status assessment report data, thus completing the final assessment process.

[0013] Preferably, for assessing the growth status of fruit trees, methods for generating a comprehensive phenotypic report by integrating leaf cover index and biomass distribution map, and determining the final monitoring results, include: Leaf cover data of crop growing areas is acquired by image acquisition equipment, and leaf distribution is recorded by high-resolution sensors to obtain a preliminary cover information dataset. Based on the preliminary coverage information dataset, the density and range characteristics of leaf coverage are extracted, and classification is performed using a preset threshold to determine the coverage level distribution results. Based on the coverage level distribution results, spatial matching processing was performed in conjunction with the biomass distribution map data collected simultaneously to obtain spatial correspondence data. If a significant discrepancy is found between leaf cover level and biomass distribution in the corresponding relationship data, the outliers are adjusted through the data correction module to obtain the corrected comprehensive distribution dataset. Based on the corrected comprehensive distribution dataset, key phenotypic features of crop growth status are extracted to generate structured feature description data; By using structured feature description data, the support vector machine algorithm is applied to classify and determine the crop growth status, and the final monitoring status result is obtained. Based on the final monitoring status results, a corresponding status assessment record is generated and stored in the database for subsequent analysis and retrieval.

[0014] The present invention also provides a fruit tree canopy phenotypic monitoring system based on multispectral images. The system is used to implement the aforementioned method and includes: a data acquisition module, an optimization module, a coverage calculation module, a biomass determination module, a growth assessment module, and a report generation module. The data acquisition module is used to acquire raw image data of farmland in arid areas through a multispectral sensor, and to process spectral channel differences using a threshold segmentation algorithm to initially distinguish fruit tree areas from soil background, thereby obtaining preliminary separation results. The optimization module is used to obtain an optimized fruit tree region based on the preliminary separation results; The coverage calculation module is used to extract the leaf pixel ratio from the optimized fruit tree area, determine the leaf coverage by calculating the pixel number ratio, and obtain the leaf coverage index. The biomass determination module is used to determine the biomass distribution map based on the leaf coverage index. The growth assessment module is used to integrate multi-temporal image data through biomass distribution maps, track change trends using time-series analysis, determine the dynamic changes of growth indicators, and obtain an assessment of the fruit tree growth status. The report generation module is used to assess the growth status of fruit trees, integrate leaf coverage indicators and biomass distribution maps to generate a comprehensive phenotypic feature report, and determine the final monitoring results.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention discloses a comprehensive technical method for monitoring fruit trees in arid regions, aiming to solve the operational problem of accurately assessing the growth status of fruit trees in complex environments. This problem involves the logical challenges of fruit tree area identification, leaf cover analysis, biomass distribution assessment, and growth dynamic tracking. This invention acquires image data using multispectral sensors, employs a threshold segmentation algorithm to initially separate fruit trees from the background, and combines this with support vector machines to optimize edge classification, eliminate environmental interference, and generate a refined fruit tree mask. Subsequently, through leaf cover index calculation and spectral intensity analysis, a biomass distribution map is constructed, and time-series analysis is performed by fusing multi-temporal images to track growth trends, ultimately generating a comprehensive phenotypic feature report. The core innovation of this invention lies in integrating multi-dimensional data analysis methods to achieve precise monitoring throughout the entire process from image processing to dynamic assessment, significantly improving the accuracy and reliability of fruit tree growth status assessment in arid regions and providing a scientific basis for agricultural management. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a method for monitoring fruit tree canopy phenotypes based on multispectral images, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a fruit tree canopy phenotypic monitoring system based on multispectral images, according to an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1 like Figure 1 As shown, this invention provides a method for monitoring fruit tree canopy phenotypic patterns based on multispectral images, the method comprising: Raw image data of farmland in arid areas were collected by multispectral sensors. Threshold segmentation algorithm was used to process spectral channel differences to initially distinguish fruit tree areas from soil background and obtain preliminary separation results. Based on the preliminary separation results, the optimized fruit tree region was obtained; The leaf pixel ratio is extracted from the optimized fruit tree area, and the leaf coverage is determined by calculating the pixel number ratio to obtain the leaf coverage index. Based on leaf cover index, determine biomass distribution map; By integrating multi-temporal image data through biomass distribution maps and using time-series analysis to track change trends, determine the dynamic changes in growth indicators, and obtain an assessment of fruit tree growth status; For the assessment of fruit tree growth status, a comprehensive phenotypic report is generated by integrating leaf cover index and biomass distribution map to determine the final monitoring results.

[0021] In this embodiment, the method of acquiring raw image data of farmland in arid areas using a multispectral sensor, and processing spectral channel differences using a threshold segmentation algorithm to initially distinguish fruit tree areas from soil background to obtain preliminary separation results includes: Use drones equipped with multispectral cameras to acquire farmland image data; Temporal and spectral feature analysis was performed on farmland image data to determine the optimal classification features, which included: Normalized Difference Vegetation Index (NDVI), Normalized Difference Green-Blue Index (NGBDI), Modified Ratio Vegetation Index (MSR), and Red Edge Band Reflectance. A decision tree classifier was constructed based on the Normalized Difference Vegetation Index (NDVI), Normalized Green-Blue Difference Index (NGBDI), Modified Ratio Vegetation Index (MSR), and red-edge band reflectance. The optimal threshold for each spectral channel was determined through repeated experiments. Based on the optimal threshold and the decision tree classifier, the separation of fruit trees from soil and weeds was achieved.

[0022] In this embodiment, the method for obtaining the optimized fruit tree region based on the preliminary separation results includes: Based on the preliminary separation results, the spectral reflectance distribution of fruit tree pixels and background pixels is obtained. If the reflectance value of the fruit tree pixels is higher than the preset threshold, it is marked as an effective tree area; otherwise, it is classified as a background area to obtain a refined fruit tree mask. Support vector machine is used to classify edge pixels in the refined fruit tree mask. For confusing pixels caused by environmental interference, the classification boundary is determined by training spectral feature vectors to obtain the optimized fruit tree region.

[0023] The method for obtaining a refined fruit tree mask involves obtaining the spectral reflectance distribution of fruit tree pixels and background pixels based on the initial separation results. If the reflectance value of the fruit tree pixels is higher than a preset threshold, they are marked as effective tree regions; otherwise, they are classified as background regions. Spectral reflectance data of fruit tree pixels and background pixels are obtained from the preliminary separation results. For each group of pixel data, the spectral reflectance value is read and recorded to obtain the initial spectral reflectance distribution dataset. For the initial spectral reflectance distribution dataset, a preset threshold is used for comparison. If the spectral reflectance value of a certain fruit tree pixel is higher than the preset threshold, it is marked as a candidate pixel of the effective region, and a preliminary set of marked pixels is obtained. Based on the initially labeled pixel set, pixels that do not reach the preset threshold are filtered and classified as background pixels, resulting in effective region pixel groups and background region pixel groups after differentiation. By performing spatial connectivity analysis on effective region pixel groups, adjacent effective region pixels are obtained to form continuous effective tree regions, resulting in a spatially optimized fruit tree region distribution. For the spatially optimized fruit tree region distribution, boundary smoothing is used to eliminate isolated pixels and irregular edges, resulting in a smoothed fruit tree region mask. Based on the smoothed fruit tree region mask, the background region pixel group is re-confirmed. If there are abnormal pixels in the background region pixel group that are adjacent to the effective tree region, they are reclassified into the effective region to obtain the final refined fruit tree mask.

[0024] Among these methods, the edge pixels in the refined fruit tree mask are classified using a support vector machine. For confusing pixels caused by environmental interference, the classification boundary is determined by training spectral feature vectors. The methods for optimizing the fruit tree region include: The process begins by acquiring raw fruit tree mask data and performing preliminary extraction on edge pixel regions. An automated segmentation method is then used to separate the target pixel set, resulting in the initial distribution of edge pixels. Specifically, by analyzing the hue and saturation distribution of sample images collected by a drone in the HSV color space, a suitable threshold range is selected to extract the green regions containing fruit tree canopies and weeds from the sample images. The extracted green region RGB images are then converted to images in Lab and HSV color space models. A Simple Linear Iterative Clustering (SLIC) superpixel segmentation algorithm is then used to pre-segment the RGB image into 250 superpixel units. Combining the superpixel segmentation information with the RGB, Lab, HSV, and grayscale images, feature vectors of the superpixel units are extracted. The feature vectors of 25% of the superpixel samples are randomly selected as the training set for an SVM classifier. The SVM classifier is then used to predict and classify all samples, achieving the segmentation of fruit tree canopies and weeds.

[0025] By extracting spectral features from the initial distribution of edge pixels, a feature vector dataset is constructed. A support vector machine is then used to train the feature vectors to determine the classification boundary model. Specifically, the classification boundary model is as follows: Weight vector and bias term: ; Decision function: ; in, The optimal weight vector, For the set of support vectors, For Lagrange multipliers, For category labels, For the first i The feature vectors of each support vector contain spectral, texture, and spatial information. The bias term of the optimal hyperplane is used to adjust the position of the classification boundary. For the first j The feature vectors of each support vector contain spectral, texture, and spatial information.

[0026] Based on the classification boundary model, the confused pixels are classified. If the spectral features of pixels related to environmental interference deviate from the preset threshold, they are classified as non-fruit tree pixels, and a preliminary classification result is obtained.

[0027] Using the preliminary classification results and the overall structure of the fruit tree mask, the edge pixels are corrected a second time to obtain the corrected pixel category distribution and determine the optimized edge range. Specifically, the boundary is determined by the Voronoi node distance, the correction area is dynamically constructed, and the root mean square error is used to quantify the correction effect. ,in, The root mean square error, To assess the total number of samples, i For the first i The index of each evaluation sample, i = 1, 2, ..., N, These are the pixel category values ​​after Voronoi diagram or morphological secondary correction. These are authentic reference values ​​obtained through manual annotation or high-precision field surveys.

[0028] By optimizing the edge range, the boundary data of the fruit tree area is reconstructed. If the boundary data deviates from the original mask, it is smoothed to obtain a refined region division. Based on the refined regional division, a final fruit tree regional distribution map is generated. Local adjustments are made to the outliers in the distribution map to obtain the final optimized result.

[0029] In this embodiment, the method for extracting the leaf pixel ratio from the optimized fruit tree area, determining the leaf coverage by calculating the pixel count ratio, and obtaining the leaf coverage index includes: By collecting image data of the fruit tree area, raw image data is obtained. The raw image data is then preprocessed to obtain the image content after preliminary cleaning.

[0030] Based on the initially cleaned image content, segmentation techniques are used to separate the leaf pixel regions, extracting leaf-related pixel data to determine the distribution range of leaf pixels. Specifically, for leaf pixel region segmentation, a color threshold-based segmentation method can be used to extract pixels within the green hue range as the leaf region. Assuming the green threshold is set to 100-200 for the G channel in the RGB values, the extracted leaf pixels are found to be mainly concentrated in the central area of ​​the image, with some pixels at the edges possibly missed due to insufficient lighting. This method can quickly locate the target area.

[0031] If the extracted leaf pixel distribution range meets the preset clarity standard, the ratio of leaf pixels to the total number of pixels in the fruit tree area is further calculated to obtain the leaf pixel ratio value. Specifically, when calculating the leaf pixel ratio value, assuming the extracted leaf pixels are 2 million and the total number of pixels is 25 million, the ratio value is 8%. By comparing it with the preset threshold of 10%, it can be preliminarily determined that the coverage is too low. This intuitive data comparison provides a basis for subsequent analysis.

[0032] By analyzing the pixel ratio values ​​of the leaves and comparing them with preset ratio thresholds, the level of leaf coverage can be determined. Specifically, when determining the final leaf coverage index, the coverage can be divided into three levels: low, medium, and high, by combining correction values ​​and level standards. Assuming 10% is a medium level, then the leaf coverage index of this farmland is medium. This grading method provides a reference for subsequent management.

[0033] If the leaf coverage level is lower than a preset minimum threshold, the image data undergoes secondary processing to obtain more refined leaf pixel data and determine a correction value for the coverage. Specifically, if the coverage level is lower than the minimum threshold, the image data can be further optimized through secondary processing, such as adjusting lighting parameters or using a higher-precision segmentation algorithm to re-extract leaf pixels. Assuming that after secondary processing, the number of leaf pixels increases to 2.5 million, and the proportion increases to 10%, the corrected coverage value is closer to the actual growth situation. This method can effectively improve the accuracy of the assessment.

[0034] Based on the revised coverage value and the grading standards, the final leaf coverage index is obtained.

[0035] In this embodiment, the method for determining the biomass distribution map based on the leaf cover index includes: Based on the leaf coverage index, the average spectral intensity value of the fruit tree area is obtained. If the average intensity value exceeds the preset threshold, it is considered a high biomass area; otherwise, it is considered a low biomass area, and a biomass distribution map is determined.

[0036] Specifically: acquire leaf cover data within the crop area, scan the target area using remote sensing imagery, extract leaf cover distribution information, and obtain a preliminary leaf cover dataset; For the preliminary dataset of leaf cover, spectral analysis tools were used to extract spectral intensity information within the crop area, calculate the spectral intensity value of each sub-region, and determine the spatial distribution data of spectral intensity. Based on the spatial distribution data of spectral intensity, the average intensity value of each sub-region is calculated. By averaging the spectral intensity of each sub-region, the distribution result of the average intensity is obtained. If the value of a certain sub-region in the distribution of average intensity is higher than the preset threshold, then the sub-region is determined to be a high biomass region; otherwise, it is determined to be a low biomass region, thus obtaining the preliminary results of biomass classification. Based on the preliminary results of biomass classification, a biomass distribution map was constructed. Geographic information system tools were used to spatially map high-biomass and low-biomass areas to determine graphical data of biomass distribution. Based on the graphical data of biomass distribution and combined with the actual boundary information of the crop area, the distribution map is subjected to boundary correction processing to obtain the final biomass distribution map data.

[0037] In this embodiment, the method for assessing the growth status of fruit trees by integrating multi-temporal image data through biomass distribution maps, using time-series analysis to track trends, and determining the dynamic changes in growth indicators includes: For the biomass distribution map, time series analysis was applied to process multi-temporal data, track the changing trends at each time point, and obtain serialized data of dynamic index values. If the serialized data of the dynamic indicator values ​​fluctuates within the preset threshold range, it is judged that the growth state is stable. If the value exceeds the preset threshold range, it is marked as an abnormal state, and a preliminary growth status assessment result is obtained. Based on the preliminary growth status assessment results and combined with the background information of crop growth status, the characteristics of key time node index changes are extracted to determine the specific classification of growth stages. By classifying the specific growth stages, the support vector machine model is used to further analyze the dynamic index values, determine the health status of crop growth, and obtain a comprehensive judgment result. After obtaining the comprehensive judgment results, for areas with abnormal conditions, the corresponding image data sources and biomass distribution information are extracted to generate targeted status assessment report data, thus completing the final assessment process.

[0038] Among them, for biomass distribution maps, the methods for processing multi-temporal data using time series analysis to track the changing trends at each time point and obtain serialized data of dynamic index values ​​include: NDVI time series smoothing: ; in, The NDVI value after smoothing at time t. These are the observations of the original NDVI time series at time t; ARIMA model: ; in, Let be the aboveground biomass at time t. For constant terms, Let the order be the autoregressive order. These are the autoregressive coefficients. This represents the historical biomass observation value at lag i. The moving average order is... Moving average coefficient For the model residuals with a lag of j periods, This represents the white noise error at the current moment.

[0039] In this embodiment, the method for assessing the growth status of fruit trees, integrating leaf cover index and biomass distribution map to generate a comprehensive phenotypic feature report, and determining the final monitoring results includes: Leaf cover data of crop growing areas is acquired by image acquisition equipment, and leaf distribution is recorded by high-resolution sensors to obtain a preliminary cover information dataset. Based on the preliminary coverage information dataset, the density and range characteristics of leaf coverage are extracted, and classification is performed using a preset threshold to determine the coverage level distribution results. Based on the coverage level distribution results, spatial matching processing was performed in conjunction with the biomass distribution map data collected simultaneously to obtain the spatial correspondence data between the two. If a significant discrepancy is found between leaf cover level and biomass distribution in the corresponding relationship data, the outliers are adjusted through the data correction module to obtain the corrected comprehensive distribution dataset. Based on the corrected comprehensive distribution dataset, key phenotypic features of crop growth status are extracted to generate structured feature description data; By using structured feature description data, the support vector machine algorithm is applied to classify and determine the crop growth status, and the final monitoring status result is obtained. Based on the final monitoring status results, a corresponding status assessment record is generated and stored in the database for subsequent analysis and retrieval.

[0040] Example 2 like Figure 2 As shown, the present invention also provides a fruit tree canopy phenotypic monitoring system based on multispectral images. The system is used to implement the method described in Embodiment 1. The system includes: a data acquisition module, an optimization module, a coverage calculation module, a biomass determination module, a growth assessment module, and a report generation module. The data acquisition module is used to acquire raw image data of farmland in arid areas through a multispectral sensor, and to process spectral channel differences using a threshold segmentation algorithm to initially distinguish fruit tree areas from soil background, thereby obtaining preliminary separation results. The optimization module is used to obtain optimized fruit tree regions based on the preliminary separation results; The coverage calculation module is used to extract the leaf pixel ratio from the optimized fruit tree area, determine the leaf coverage by calculating the pixel number ratio, and obtain the leaf coverage index. The biomass determination module is used to determine the biomass distribution map based on the leaf cover index. The growth assessment module is used to integrate multi-temporal image data through biomass distribution maps, track change trends using time-series analysis, determine the dynamic changes of growth indicators, and obtain an assessment of the fruit tree growth status. The report generation module is used to assess the growth status of fruit trees, integrate leaf coverage indicators and biomass distribution maps to generate a comprehensive phenotypic report, and determine the final monitoring results.

[0041] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for monitoring fruit tree canopy phenotypic patterns based on multispectral images, characterized in that, The method includes: Raw image data of farmland in arid areas were collected by multispectral sensors. Threshold segmentation algorithm was used to process spectral channel differences to initially distinguish fruit tree areas from soil background and obtain preliminary separation results. Based on the preliminary separation results, the optimized fruit tree region was obtained; The leaf pixel ratio is extracted from the optimized fruit tree area, and the leaf coverage is determined by calculating the pixel number ratio to obtain the leaf coverage index. Based on leaf cover index, determine biomass distribution map; By integrating multi-temporal image data through biomass distribution maps and using time-series analysis to track change trends, determine the dynamic changes in growth indicators, and obtain an assessment of fruit tree growth status; For the assessment of fruit tree growth status, a comprehensive phenotypic report is generated by integrating leaf cover index and biomass distribution map to determine the final monitoring results.

2. The method according to claim 1, characterized in that, Raw image data of farmland in arid regions was acquired using multispectral sensors. Threshold segmentation algorithms were then used to process spectral channel differences to initially distinguish fruit tree areas from the soil background, yielding preliminary separation results. The methods included: Temporal and spectral feature analysis was performed on farmland image data to determine the optimal classification features, which included: Normalized Difference Vegetation Index (NDVI), Normalized Difference Green-Blue Index (NGBDI), Modified Ratio Vegetation Index (MSR), and Red Edge Band Reflectance. A decision tree classifier was constructed based on the Normalized Difference Vegetation Index (NDVI), the Normalized Green-Blue Difference Index (NGBDI), the Modified Ratio Vegetation Index (MSR), and the red-edge band reflectance. The optimal threshold for each spectral channel was determined through repeated experiments. Based on the optimal threshold and the decision tree classifier, the separation of fruit trees from soil and weeds was achieved.

3. The method according to claim 1, characterized in that, Based on the preliminary separation results, methods for obtaining optimized fruit tree regions include: Based on the preliminary separation results, the spectral reflectance distribution of fruit tree pixels and background pixels is obtained. If the reflectance value of the fruit tree pixels is higher than the preset threshold, it is marked as an effective tree area; otherwise, it is classified as a background area to obtain a refined fruit tree mask. Support vector machine is used to classify edge pixels in the refined fruit tree mask. For confusing pixels caused by environmental interference, the classification boundary is determined by training spectral feature vectors to obtain the optimized fruit tree region.

4. The method according to claim 1, characterized in that, Methods for extracting leaf pixel ratios from optimized fruit tree areas and determining leaf coverage by calculating the pixel count ratio to obtain leaf coverage indicators include: By collecting image data of the fruit tree area, raw image data is obtained, and the raw image data is preprocessed to obtain the image content after preliminary cleaning. Based on the image content after preliminary cleaning, the leaf pixel region is separated using segmentation technology, and the pixel data related to the leaf is extracted to determine the distribution range of the leaf pixels. If the extracted leaf pixel distribution range meets the preset clarity standard, then the ratio of the number of leaf pixels to the total number of pixels in the fruit tree area is further calculated to obtain the leaf pixel ratio value. By analyzing the pixel ratio values ​​of the leaves and comparing them with preset ratio thresholds, the level of leaf coverage is determined. If the leaf coverage level is lower than the preset minimum threshold, the image data will be processed again to obtain more refined leaf pixel data and determine the correction value of the coverage. Based on the revised coverage value and the grading standards, the final leaf coverage index is obtained.

5. The method according to claim 1, characterized in that, Methods for determining biomass distribution maps based on leaf cover indices include: Based on the leaf coverage index, the average spectral intensity value of the fruit tree area is obtained. If the average intensity value exceeds the preset threshold, it is considered a high biomass area; otherwise, it is considered a low biomass area, and a biomass distribution map is determined.

6. The method according to claim 1, characterized in that, Methods for assessing fruit tree growth status include integrating multi-temporal image data using biomass distribution maps, employing time-series analysis to track trends, and determining dynamic changes in growth indicators. For the biomass distribution map, time series analysis was applied to process multi-temporal data, track the changing trends at each time point, and obtain serialized data of dynamic index values. If the serialized data of the dynamic indicator values ​​fluctuates within the preset threshold range, it is judged that the growth state is stable. If the value exceeds the preset threshold range, it is marked as an abnormal state, and a preliminary growth status assessment result is obtained. Based on the preliminary growth status assessment results and combined with the background information of crop growth status, the characteristics of key time node index changes are extracted to determine the specific classification of growth stages. By classifying the specific growth stages, the support vector machine model is used to further analyze the dynamic index values, determine the health status of crop growth, and obtain a comprehensive judgment result. After obtaining the comprehensive judgment results, for areas with abnormal conditions, the corresponding image data sources and biomass distribution information are extracted to generate targeted status assessment report data, thus completing the final assessment process.

7. The method according to claim 1, characterized in that, For assessing the growth status of fruit trees, methods for generating a comprehensive phenotypic report by integrating leaf cover index and biomass distribution map, and determining the final monitoring results, include: Leaf cover data of crop growing areas is acquired by image acquisition equipment, and leaf distribution is recorded by high-resolution sensors to obtain a preliminary cover information dataset. Based on the preliminary coverage information dataset, the density and range characteristics of leaf coverage are extracted, and classification is performed using a preset threshold to determine the coverage level distribution results. Based on the coverage level distribution results, spatial matching processing was performed in conjunction with the biomass distribution map data collected simultaneously to obtain spatial correspondence data. If a significant discrepancy is found between leaf cover level and biomass distribution in the corresponding relationship data, the outliers are adjusted through the data correction module to obtain the corrected comprehensive distribution dataset. Based on the corrected comprehensive distribution dataset, key phenotypic features of crop growth status are extracted to generate structured feature description data; By using structured feature description data, the support vector machine algorithm is applied to classify and determine the crop growth status, and the final monitoring status result is obtained. Based on the final monitoring status results, a corresponding status assessment record is generated and stored in the database for subsequent analysis and retrieval.

8. A fruit tree canopy phenotypic monitoring system based on multispectral images, the system being used to implement the method described in any one of claims 1-7, characterized in that, The system includes: a data acquisition module, an optimization module, a coverage calculation module, a biomass determination module, a growth assessment module, and a report generation module; The data acquisition module is used to acquire raw image data of farmland in arid areas through a multispectral sensor, and to process spectral channel differences using a threshold segmentation algorithm to initially distinguish fruit tree areas from soil background, thereby obtaining preliminary separation results. The optimization module is used to obtain an optimized fruit tree region based on the preliminary separation results; The coverage calculation module is used to extract the leaf pixel ratio from the optimized fruit tree area, determine the leaf coverage by calculating the pixel number ratio, and obtain the leaf coverage index. The biomass determination module is used to determine the biomass distribution map based on the leaf coverage index. The growth assessment module is used to integrate multi-temporal image data through biomass distribution maps, track change trends using time-series analysis, determine the dynamic changes of growth indicators, and obtain an assessment of the fruit tree growth status. The report generation module is used to assess the growth status of fruit trees, integrate leaf coverage indicators and biomass distribution maps to generate a comprehensive phenotypic feature report, and determine the final monitoring results.

Images