Crop phenotype recognition method and system based on computer vision

Abstract

The application provides a crop phenotype recognition method and system based on computer vision, relates to the technical field of computer vision, acquires time-series crop visual sequences of crops planted in saline-alkali land and corresponding time-series saline-alkali land environment visual sequences, extracts visual images of the same time node in the time-series crop visual sequences and the time-series saline-alkali land environment visual sequences, establishes a space position corresponding relationship between the crop visual images and the saline-alkali land environment visual images at the same time node, and obtains a time-correlated visual image pair set; performs feature interaction extraction processing on the image pairs in the time-correlated visual image set to generate crop environment interaction features; based on the time variation law of the time-series crop visual sequences, performs evolution trajectory modeling processing on the crop environment interaction features of each time node to construct the evolution trajectory of the crop environment interaction features; and generates a crop phenotype recognition result according to the evolution trajectory and outputs a report containing associated information, so that the crop phenotype in the saline-alkali land can be accurately recognized.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and more specifically, to a method and system for crop phenotypic recognition based on computer vision. Background Technology

[0002] In agriculture, a precise understanding of crop phenotypic characteristics is crucial for crop breeding, cultivation management, and yield prediction. Crop phenotype refers to the sum of morphological, structural, and physiological / biochemical characteristics exhibited by a crop under the interaction of genotype and environment. Traditional methods for crop phenotypic identification primarily rely on manual observation and measurement. These methods are not only inefficient but also susceptible to subjective influences, making it difficult to guarantee the accuracy and consistency of the identification results.

[0003] With the development of computer vision technology, image-based crop phenotypic recognition methods have gradually emerged. However, most existing computer vision-based crop phenotypic recognition methods only focus on the visual features of the crop itself, neglecting the influence of the crop's growth environment on its phenotypic characteristics. Especially in the special planting environment of saline-alkali land, factors such as soil salinity, moisture conditions, and irrigation conditions have a significant impact on crop growth and development, thus affecting the crop's phenotypic characteristics. Therefore, phenotypic recognition based solely on the crop's visual features cannot comprehensively and accurately reflect the true phenotypic characteristics of crops in saline-alkali land environments, and it is difficult to meet the needs of modern agriculture for precise crop phenotypic recognition. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a crop phenotypic recognition method based on computer vision, the method comprising:

[0005] The method acquires a temporal visual sequence of crops and a corresponding temporal visual sequence of the saline-alkali land environment within a planting area. The temporal visual sequence of crops includes visual images of leaves, stems, and roots at different growth stages, and the visual images are arranged in chronological order. The temporal visual sequence of the saline-alkali land environment includes visual images of the soil surface, soil profile, and irrigation area at corresponding times within the crop growth area, and the visual images correspond to the temporal order of the temporal visual sequence of crops.

[0006] By extracting visual images from the same time point in the time-series crop visual sequence and the time-series saline-alkali land environment visual sequence, the spatial positional correspondence between crop visual images and saline-alkali land environment visual images at the same time point is established, resulting in a set of time-related visual image pairs.

[0007] Perform feature interaction extraction processing on each temporally associated visual image pair in the set to generate crop-environment interaction features;

[0008] Based on the temporal variation patterns of crop visual sequences, we perform evolutionary trajectory modeling on the crop-environment interaction features at each time point, analyze the differences in changes of crop-environment interaction features at adjacent time points, and construct the evolutionary trajectory of crop-environment interaction features.

[0009] Crop phenotypic recognition results are generated based on the evolutionary trajectory of crop-environment interaction characteristics. Based on these results, a phenotypic recognition report containing information on the correlation between phenotypic features and saline-alkali land environment is output. The crop phenotypic recognition results include crop growth status characteristics, morphological and structural characteristics, and physiological response characteristics to the saline-alkali land environment.

[0010] Furthermore, embodiments of the present invention also provide a crop phenotypic recognition system based on computer vision, characterized in that it includes:

[0011] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to perform the above-described computer vision-based crop phenotypic recognition method by executing the machine-executable instructions.

[0012] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions stored in a computer-readable storage medium, the processor of the computer vision-based crop phenotyping system reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer vision-based crop phenotyping system to perform the above-described computer vision-based crop phenotyping method.

[0013] Based on the above, by acquiring the temporal visual sequences of crops and the corresponding temporal visual sequences of the saline-alkali land environment within the planting area, a spatial correspondence between crop visual images and saline-alkali land environment visual images at the same time node is established, forming a set of temporally correlated visual image pairs. This allows for precise location of the spatial relationship between crops and the environment. Feature interaction extraction processing is performed on the temporally correlated visual image pairs to generate crop-environment interaction features, fully considering the mutual influence between crop organ features and environmental features, making the recognition results more reflective of the true state of crops in the saline-alkali land environment. Based on the temporal variation patterns, an evolutionary trajectory of crop-environment interaction features is constructed, enabling dynamic analysis of the trend of crop phenotype changes with time and environment. Finally, based on the evolutionary trajectory, crop phenotype recognition results are generated and a report containing correlation information is output, achieving comprehensive and accurate identification of crop phenotypes and significantly improving the accuracy of crop phenotype recognition. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the execution flow of the crop phenotypic recognition method based on computer vision provided in an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of exemplary hardware and software components of a computer vision-based crop phenotypic recognition system provided in an embodiment of the present invention. Detailed Implementation

[0016] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a crop phenotypic recognition method based on computer vision according to an embodiment of the present invention. The following is a detailed description of the crop phenotypic recognition method based on computer vision.

[0017] Step S110: Obtain the temporal crop visual sequence and the corresponding temporal saline-alkali land environment visual sequence within the saline-alkali land planting area. The temporal crop visual sequence includes leaf visual images, stem visual images, and root visual images of the crop at different growth cycles, and each visual image is arranged in chronological order of growth time. The temporal saline-alkali land environment visual sequence includes soil surface visual images, soil profile visual images, and irrigation area visual images of the crop growth area at the corresponding time, and each visual image corresponds to the temporal order of the temporal crop visual sequence.

[0018] In this embodiment, wheat crops in a specific saline-alkali land planting area are used as the research object to acquire the aforementioned temporal visual sequence. First, it is necessary to clarify that the goal is to acquire visual data that comprehensively reflects the changes in the crop itself and its surrounding environment during the wheat's growth in the saline-alkali land environment. This data will serve as the basis for subsequent phenotypic recognition. To achieve this goal, a series of operations, from determining the planting area to executing the image acquisition plan, need to be systematically carried out.

[0019] Step S111: Determine the crop planting area and environmental monitoring area of ​​the saline-alkali land planting area. After adjusting the environmental monitoring area to fully cover the crop planting area, adjust the acquisition parameters of the relevant visual acquisition equipment to adapt the acquisition parameters to the light conditions of the saline-alkali land environment and the height characteristics of crop growth. The acquisition parameters include acquisition resolution, acquisition frame rate and exposure time.

[0020] In this embodiment, for the aforementioned wheat planting area, the actual planting area of ​​the wheat is first determined through on-site surveys combined with satellite remote sensing images. This actual planting area presents as an irregular polygonal region. The environmental monitoring range is initially set as a rectangular area slightly larger than the crop planting area. Through multiple adjustments to the boundary, it is ensured that the environmental monitoring range can completely cover the crop planting area, avoiding any omissions in environmental data collection.

[0021] Next, the acquisition parameters of the visual acquisition equipment were adjusted. Considering the strong light and significant variations in light duration in the saline-alkali soil environment, as well as the gradual increase in height of wheat from seedling to maturity, the acquisition resolution was set to a value that could clearly distinguish the texture of wheat leaves and the details of the soil surface, ensuring that the image contained sufficient feature information. The acquisition frame rate was set to a low value based on the slow growth of wheat to avoid unnecessary data redundancy. The exposure time was dynamically adjusted according to the light intensity at different times of day. For example, during the intense midday light, the exposure time was shortened to prevent overexposure; during the weaker light in the early morning or evening, the exposure time was appropriately extended to ensure moderate image brightness and clear presentation of the visual characteristics of the crop and its environment.

[0022] Step S112: Construct a visual acquisition time plan, determine the acquisition interval according to the crop growth cycle, and first acquire visual images of the crop within the crop planting area at each acquisition time point. During acquisition, take pictures from multiple angles for the leaf parts, stem parts and root parts of the crop. Through multi-angle shooting, the visual image of each part can present the shape and color characteristics of that part.

[0023] For wheat, its growth cycle includes stages such as seedling stage, tillering stage, jointing stage, heading stage, grain-filling stage, and maturity stage. The harvesting interval is determined based on the different growth rates at each stage. During the relatively rapid growth stages of jointing and heading, the harvesting interval is set to a shorter interval, such as once every three days; during the relatively slow growth stages of seedling and maturity, the harvesting interval is set to a longer interval, such as once every seven days.

[0024] At each collection point, visual images of the crop were first collected according to the predetermined plan. For the leaves, images were taken from different horizontal and vertical angles around the wheat plant, such as from the front, left front, right front, and above the plant, to comprehensively capture the leaf's extension shape, curling degree, and color distribution, including whether the leaves showed yellowing, spots, or other color changes. For the stems, images were also taken from multiple angles, focusing on the stem's thickness, internode length, and whether there was a tendency to lodging, while also recording color changes on the stem surface, such as fading or disease. For the roots, since wheat roots grow in the soil, special collection methods were required, such as using transparent root observation tubes or digging profiles. The roots were then photographed from different angles to show the root distribution range, number of branches, and root color characteristics, including whether the roots showed browning or rot.

[0025] Step S113: After completing the acquisition of crop visual images at the same acquisition time point, immediately acquire the environmental visual images of saline-alkali land within the corresponding environmental monitoring range. During acquisition, fixed-point shooting is carried out on the soil surface, soil profile and irrigation area respectively. Through fixed-point shooting, the visual images of each environmental area can fully present the texture and structural features of the environmental area.

[0026] After acquiring visual images of wheat crops, visual images of the saline-alkali land environment were then acquired. For the soil surface, multiple representative fixed locations were selected for imaging. These locations were evenly distributed within the environmental monitoring range to reflect the overall condition of the soil surface. During imaging, the lens was ensured to be perpendicular to the soil surface to fully present the texture and structural features of the soil surface, such as cracking, distribution of saline-alkali leaching, vegetation cover, and flatness.

[0027] For soil profiles, multiple soil profile pits are excavated within the environmental monitoring area. The depth of each pit is determined based on the main distribution depth of the wheat root system, generally excavating to a depth sufficient to observe the soil stratification structure of the main wheat root activity area. Fixed-point photography is taken on the side of each pit to clearly show the soil stratification distribution, color differences between layers, texture (such as the proportion of sand, loam, and clay), and structural features such as the presence of hard soil layers.

[0028] For irrigated areas, fixed points are set up at key locations such as the outlet of the irrigation system, the area through which the water flows, and the area where water accumulates after irrigation. During filming, the distribution of irrigation water, the flow rate, the boundary of the irrigated area, and the degree of soil saturation are recorded to reflect the impact of irrigation on the saline-alkali land environment.

[0029] Step S114: Classify the leaf visual images, stem visual images, and root visual images obtained at each collection time point into crop visual image groups for that collection time point. Arrange all crop visual image groups at all time points in chronological order of collection time to form a time-series crop visual sequence. Record the specific collection time information corresponding to each crop visual image group during the arrangement process.

[0030] After each data collection point, the captured visual images of wheat leaves, stems, and roots are organized and grouped into a single crop visual image set. For example, at a certain data collection point, if 5 leaf images, 4 stem images, and 3 root images are collected, these images are combined to form the crop visual image set for that time point.

[0031] Then, according to the chronological order of collection time, all crop visual image groups from different collection time points are arranged sequentially. For example, the first crop visual image group is placed at the beginning of the sequence, the second group follows, and so on. During the arrangement process, specific collection time information, such as year, month, day, hour, and minute, is added to each crop visual image group to facilitate subsequent time correlation analysis. Through the above operations, a time-series crop visual sequence is finally formed, which completely records the crop visual information of wheat at different growth time points.

[0032] Step S115: Classify the visual images of soil surface, soil profile and irrigation area obtained at each collection time point into a visual image group of saline-alkali land environment at that collection time point, and arrange all the visual image groups of saline-alkali land environment at all time points in chronological order of collection time to form an initial temporal visual sequence of saline-alkali land environment.

[0033] Similar to the processing of crop visual image groups, at each acquisition time point, the captured visual images of the soil surface, soil profile, and irrigated area are categorized into a visual image group for the saline-alkali land environment. For example, if 6 soil surface images, 3 soil profile images, and 4 irrigated area images are acquired at a certain acquisition time point, they are combined into a visual image group for the saline-alkali land environment at that time point.

[0034] Next, in chronological order of data collection, the visual images of the saline-alkali land environment at all time points were arranged sequentially to form an initial temporal visual sequence of the saline-alkali land environment. This sequence initially reflects the changes in the environment of the wheat-growing area over time.

[0035] Step S116: Compare the acquisition time information of the time-series crop visual sequence with the initial time-series saline-alkali land environment visual sequence, and adjust the arrangement order of each saline-alkali land environment visual image group in the initial time-series saline-alkali land environment visual sequence so that the acquisition time of each saline-alkali land environment visual image group is completely consistent with the acquisition time of the corresponding crop visual image group in the time-series crop visual sequence.

[0036] In actual data collection, there may be slight differences in the acquisition time between crop visual images and environmental visual images at the same point in time, or the image acquisition order may be reversed due to equipment malfunctions or other reasons. Therefore, it is necessary to compare the acquisition time information of each image group in the time-series crop visual sequence and the initial time-series saline-alkali land environmental visual sequence.

[0037] For example, in a time-series crop visual image sequence, the acquisition time of a certain crop visual image group is a specific moment, while the acquisition time of the corresponding saline-alkali land environment visual image group in the initial time-series saline-alkali land environment visual sequence may be earlier or later than that moment, or the order may be incorrect. In this case, it is necessary to adjust the arrangement order of the saline-alkali land environment visual image groups according to the acquisition time of the crop visual image groups to ensure that the two correspond one-to-one in time, so that the acquisition time of each saline-alkali land environment visual image group is completely consistent with the acquisition time of the corresponding crop visual image group.

[0038] Step S117: Confirm that each group of saline-alkali land environment visual images in the adjusted time-series saline-alkali land environment visual sequence includes a soil surface visual image, a soil profile visual image, and an irrigation area visual image, and that the clarity and completeness of each visual image meet the requirements of subsequent processing, and finally obtain a time-series saline-alkali land environment visual sequence that corresponds to the time sequence of the time-series crop visual sequence.

[0039] After adjusting the order, the adjusted temporal visual sequence of saline-alkali land environment was checked one by one. It was confirmed that each set of saline-alkali land environmental visual images contained all three types of images: soil surface visual images, soil profile visual images, and irrigation area visual images; none could be missing. If any image set was found to be missing a certain type of image, it was necessary to re-acquire or supplement it from other backup data.

[0040] Simultaneously, the sharpness of each visual image is checked by observing whether details in the image are clearly distinguishable, such as the texture of the soil surface, the boundaries of soil profile layers, and the details of water flow in the irrigation area, to ensure that the image is free from blurring, ghosting, or other issues that would affect subsequent feature extraction. Regarding completeness, the images are checked for excessive edge cropping or missing key areas. Only when all image groups meet the above requirements can the final visual sequence of the saline-alkali land environment be determined, ensuring a precise temporal correspondence with the visual sequence of the crops.

[0041] Step S120: By extracting visual images from the same time node in the time-series crop visual sequence and the time-series saline-alkali land environment visual sequence, the spatial positional correspondence between the crop visual image and the saline-alkali land environment visual image at the same time node is established, and a set of time-related visual image pairs is obtained.

[0042] After obtaining the temporal crop visual sequence and the temporal saline-alkali land environment visual sequence, it is necessary to correlate the two in the time dimension and further establish the spatial location correspondence.

[0043] Step S121: Extract the acquisition time nodes corresponding to all crop visual image groups from the time-series crop visual sequence and the time-series saline-alkali land environment visual sequence to form a crop time node list and an environment time node list.

[0044] In this embodiment, for the aforementioned temporal crop visual sequence of wheat, each crop visual image group is traversed, and their respective corresponding acquisition time nodes are extracted. These time nodes are accurate to the specific acquisition moment, and the above time nodes are arranged in chronological order to form a crop time node list. Similarly, each saline-alkali land environment visual image group in the temporal saline-alkali land environment visual sequence is processed to extract its acquisition time nodes, forming an environment time node list. For example, the crop time node list may contain time nodes such as "T1, T2, T3...Tn", and the environment time node list also contains similar time nodes, but there may be slight differences or repetitions in some time nodes.

[0045] Step S122: Match the crop time node list with the environmental time node list, filter out time nodes with completely identical collection time information, determine the filtered time nodes as associated time nodes, and form an associated time node list.

[0046] The crop time node list and the environmental time node list are compared one by one. Time nodes with identical collection time information (year, month, day, hour, minute) in both lists are selected as associated time nodes. For example, if T2 in the crop time node list and T2 in the environmental time node list have the exact same collection time, then T2 is identified as an associated time node. Through the above selection process, a list of associated time nodes is obtained.

[0047] Step S123: For the first associated time node in the associated time node list, extract the crop visual image group corresponding to the associated time node from the time-series crop visual sequence. The crop visual image group includes leaf visual images, stem visual images and root visual images.

[0048] Taking the first associated time node in the associated time node list, such as T1, as an example, find the crop visual image group corresponding to T1 from the time-series crop visual sequence. This crop visual image group is the collection of all wheat crop images collected at time T1, including multiple leaf visual images, stem visual images, and root visual images. These images comprehensively reflect the growth status of wheat at time T1.

[0049] Step S124: Extract the visual image group of saline-alkali land environment corresponding to the associated time node from the temporal visual sequence of saline-alkali land environment. The visual image group of saline-alkali land environment includes visual images of soil surface, visual images of soil profile and visual images of irrigation area.

[0050] Similarly, for the associated time node T1, the corresponding visual image set of the saline-alkali land environment was extracted from the time-series visual sequence of the saline-alkali land environment. This image set includes visual images of the soil surface, soil profile, and irrigation area collected at time T1. These images reflect the condition of the wheat growth environment at time T1.

[0051] Step S125: Perform spatial coordinate calibration on the crop visual image group and the saline-alkali land environment visual image group under the associated time node. Using the fixed marker in the crop planting area as a reference, determine the spatial coordinates of the fixed marker in the crop visual image and the saline-alkali land environment visual image respectively. Make the spatial coordinates of the fixed marker in the two sets of images consistent through coordinate transformation.

[0052] In this embodiment, several fixed markers within the crop planting area are selected as references for spatial coordinate calibration, such as metal poles buried in the soil and fixed cement piles. These markers can be clearly captured in both crop visual images and saline-alkali land environment visual images.

[0053] First, in the visual images of crops, these fixed markers are located using image recognition technology, and their coordinate positions in the image coordinate system are determined. For example, for a certain metal marker, the pixel coordinates of its top and bottom in the crop leaf image are determined. Similarly, in the visual images of saline-alkali land environments, the same metal markers are found, and their pixel coordinates in the image coordinate system are determined.

[0054] Then, based on the actual positional relationships of these markers in real space, a transformation relationship between the image coordinate system and the real space coordinate system is established. Using a coordinate transformation algorithm, the coordinates of markers in the crop visual image and the saline-alkali land environment visual image are transformed to the same real space coordinate system. The image scaling, rotation angle, and translation parameters are adjusted to ensure that the spatial coordinates of all fixed markers in the two sets of images remain consistent.

[0055] Step S126: Based on the calibrated spatial coordinates, establish the spatial positional correspondence between the leaf visual image and the soil surface visual image, the spatial positional correspondence between the stem visual image and the soil profile visual image, and the spatial positional correspondence between the root visual image and the irrigation area visual image at the associated time node, forming a time-related visual image pair for the associated time node.

[0056] Based on the calibrated spatial coordinates, the spatial correspondence between various types of images is established. For leaf visual images and soil surface visual images, since the leaves grow on the soil surface, the calibrated spatial coordinates can be used to determine the projection area of ​​the leaf on the soil surface in each leaf visual image, thereby finding the position range of the corresponding soil surface visual image, and establishing the spatial correspondence between the corresponding areas in the leaf visual image and the soil surface visual image.

[0057] A spatial correspondence between visual images of stems and visual images of soil profiles is established, considering that stems grow from the soil surface and their root systems are distributed within the soil profile. By determining the position of the stem in the spatial coordinate system, the corresponding soil profile region can be identified, and thus the visual image of the stem can be associated with the visual image of the soil profile in that region, reflecting the relationship between stem growth and soil profile structure.

[0058] A spatial correspondence was established between root visual images and irrigation area visual images. Based on the distribution of roots in the soil and the spatial range of the irrigation area, the corresponding irrigation area was identified, and the root visual images were correlated with the visual images of this irrigation area to analyze the impact of irrigation on root growth. By establishing these three spatial correspondences, the crop visual images and saline-alkali land environment visual images at the associated time points were combined to form time-correlated visual image pairs.

[0059] Step S127: Process all associated time nodes in the associated time node list in sequence, generate a time-related visual image pair for each associated time node, integrate all the generated time-related visual image pairs to obtain a set of time-related visual image pairs. Each element in the set of time-related visual image pairs contains associated time node information, crop visual image, saline-alkali land environment visual image and the spatial positional correspondence between the two.

[0060] Following the same method used to process the first associated time node, the other associated time nodes in the associated time node list are processed sequentially. For each associated time node, steps such as extracting the corresponding image group from the time-series crop visual sequence and the time-series saline-alkali land environment visual sequence, performing spatial coordinate calibration, and establishing spatial location correspondence are performed to generate corresponding time-related visual image pairs.

[0061] After all temporally associated visual image pairs for all associated time points have been generated, they are integrated together to form a temporally associated visual image pair set. Each element in this set is a complete temporally associated visual image pair, containing information about the associated time point, the corresponding crop visual image, the saline-alkali land environment visual image, and the spatial correspondence between the two, thus preparing for subsequent feature interaction extraction processing.

[0062] Step S130: Perform feature interaction extraction processing on each time-related visual image pair in the time-related visual image pair set to generate crop-environment interaction features.

[0063] After the time-related visual image pair set is established, in-depth feature extraction and interaction mapping are required for each image pair to obtain comprehensive features that can reflect the interaction between crops and the environment.

[0064] Step S131: Extract the first time-related visual image pair from the set of time-related visual image pairs. The time-related visual image pair includes crop visual images, saline-alkali land environment visual images and the spatial location correspondence between the two.

[0065] In this embodiment, the first time-related visual image pair is extracted from the aforementioned set of time-related visual image pairs. This image pair corresponds to the earliest associated time node and includes visual images of wheat crops (including leaf, stem, and root images), visual images of the saline-alkali land environment (including soil surface, soil profile, and irrigation area images) at that time node, and the established spatial correspondence between them.

[0066] Step S132: Perform organ region segmentation on the crop visual image. Based on the color differences and morphological contours of the crop organs, divide the crop visual image into leaf region, stem region and root region. Visual features are extracted separately for each region.

[0067] Organ region segmentation is performed on the crop visual image in the first temporally correlated visual image pair. Taking leaf region segmentation as an example, wheat leaves are usually green (with varying shades of green at different growth stages), which is significantly different from the stem and background soil color. Additionally, leaves have a specific elongated shape. Using image segmentation algorithms, color thresholding combined with morphological operations (such as erosion, dilation, and region filling) is employed to separate the leaves from the entire crop visual image, forming leaf regions.

[0068] Similarly, the stem region is segmented based on its color (usually green or yellowish-green, potentially turning yellow upon maturity) and shape (cylindrical, upright growth) compared to the leaves and roots. The root region is segmented based on its distribution pattern in the soil, color (usually white or light yellow, turning brown with age), and contrast with the soil background. Through this segmentation process, the crop visual image is clearly divided into leaf, stem, and root regions, each independent of the others, facilitating the subsequent extraction of visual features individually.

[0069] Step S133: For the leaf area, extract the color distribution features, texture arrangement features and edge contour features of the leaf area. The color distribution features reflect the distribution ratio of different colors in the leaf area, the texture arrangement features reflect the arrangement pattern of the texture on the leaf surface, and the edge contour features reflect the shape features of the leaf contour.

[0070] For the segmented leaf regions, color distribution features are first extracted. The image of the leaf region is converted to a suitable color space (such as the HSV color space), and then the distribution of pixel values ​​in different color channels (e.g., H channel represents hue, S channel represents saturation, and V channel represents lightness) is statistically analyzed. The proportion of pixels in each color range to the total number of pixels in the leaf region is calculated to obtain the color distribution features. For example, the distribution ratio of green hue in the H channel, the distribution ratio of yellow hue, etc., are statistically analyzed. These ratios reflect the color condition of the leaf, such as whether it is deficient in nutrients or whether it is diseased.

[0071] Texture arrangement features are extracted by analyzing the spatial distribution of pixel grayscale values ​​on the leaf surface. Texture analysis algorithms (such as gray-level co-occurrence matrix, LBP operator, etc.) are used to calculate texture feature parameters of the leaf region at different directions and distances, such as energy, entropy, contrast, correlation, etc. These parameters reflect the arrangement rules of the leaf surface texture, such as coarseness, density, and regularity. For example, the surface texture of healthy wheat leaves may be relatively uniform and regular, while the texture of leaves attacked by pests and diseases may become messy and disordered.

[0072] Edge contour feature extraction involves using an edge detection algorithm (such as the Canny edge detection algorithm) to find the edge pixels in the blade region. These edge pixels are then fitted to obtain the blade's contour curve. Based on this contour curve, shape feature parameters such as length, area, perimeter, concavity / convexity, and curvature variation are calculated. These parameters reflect the overall shape of the blade, such as its aspect ratio, presence of notches, and curling.

[0073] Step S134: For the stem region, extract the color uniformity feature, texture density feature and morphological structure feature of the stem region. The color uniformity feature reflects the uniformity of the color in the stem region, the texture density feature reflects the density of the texture on the stem surface, and the morphological structure feature reflects the thickness and curvature of the stem.

[0074] For the stem region, the color uniformity feature is extracted by calculating statistical measures such as the standard deviation and variance of the color distribution in the stem region. If the stem color distribution is relatively concentrated and the standard deviation and variance are small, the color uniformity is high, indicating that the stem growth is relatively stable; conversely, the color uniformity is low, which may indicate problems such as unbalanced nutrition or disease.

[0075] Texture density features are extracted to analyze the quantity and distribution density of textures on the stem surface. The density of the texture is measured by calculating parameters such as the number of texture feature points per unit area or the total length of texture lines. For example, the distribution density of pubescence on the surface of wheat stems and the texture density at the internodes can reflect the maturity and health of the stems.

[0076] The extraction of morphological and structural features mainly focuses on the thickness and curvature of the stem. Multiple feature points are selected on the contour of the stem region to calculate parameters such as stem diameter (e.g., diameter values ​​at different heights), stem height, and the angle between the stem axis and the vertical direction. The diameter reflects the stem's robustness, and the angle reflects its curvature; these parameters are crucial for assessing the stem's resistance to lodging and its growth status.

[0077] Step S135: For the root system region, extract the color depth features, texture direction features, and branch distribution features of the root system region. The color depth features reflect the color depth changes of the root system region, the texture direction features reflect the extension direction of the root texture, and the branch distribution features reflect the distribution of root branches.

[0078] The color depth of the root system is represented by calculating the average gray value of the root system image or the average value of a specific color channel. The lower the average gray value (in a grayscale image), the darker the root color, which may reflect the degree of root aging or vitality; conversely, a lighter color may indicate that the roots are more tender.

[0079] The extraction of texture orientation features is achieved by analyzing the main extension direction of root texture. Methods such as directional filters or Hough transforms are used to detect the energy distribution of root texture in different directions, determining the main orientation of the texture, such as transverse, longitudinal, and oblique. These orientation features reflect the direction and trend of root growth and are related to the distribution of moisture and nutrients in the soil environment.

[0080] The branch distribution characteristics are extracted, and parameters such as the number, length, angle, and spacing between branches within the root system area are statistically analyzed. By extracting the skeleton and detecting branch points in the root system image, the topological structure information of the root system is obtained, thereby analyzing the uniformity, density, and growth status of root branches. These characteristics are crucial for assessing the root system's absorption capacity and adaptability to the soil environment.

[0081] Step S136: Perform environmental region segmentation on the visual image of the saline-alkali land environment. Based on the texture differences and structural features of the environmental regions, divide the visual image of the saline-alkali land environment into soil surface region, soil profile region and irrigation region. Extract environmental visual features for each region separately.

[0082] Environmental region segmentation is performed on visual images of saline-alkali land. For soil surface areas, segmentation is based on differences in surface texture (e.g., cracked texture, granular texture) and structure (e.g., smoothness, presence of saline crust) compared to soil profiles and irrigated areas. For example, cracked textures on the soil surface have specific geometric shapes and distribution patterns, which can be segmented using texture feature recognition combined with region growing algorithms.

[0083] Soil profile areas are segmented based on their hierarchical structural characteristics. Different soil layers differ in color, texture, and compaction, forming a distinct horizontal stratification structure. These structural characteristics can be used to separate soil profile areas from environmental visual images.

[0084] The irrigation area was segmented based on the color difference between the irrigation water and the soil (water is usually darker or has specular reflective properties) and the shape of the irrigation area (e.g., distributed along irrigation canals or in patches). Through this segmentation process, the visual image of the saline-alkali land environment was divided into three independent environmental regions.

[0085] Step S137: For the soil surface area, extract the color change characteristics, texture roughness characteristics, and crack distribution characteristics of the soil surface area; for the soil profile area, extract the layer distribution characteristics, color transition characteristics, and structural density characteristics of the soil profile area; for the irrigation area, extract the water distribution characteristics, color wetness characteristics, and boundary clarity characteristics of the irrigation area.

[0086] For soil surface areas, color variation features are extracted by analyzing color differences at different locations on the soil surface, calculating statistical measures such as the mean and variance of each channel in the color space to reflect the spatial unevenness of soil salinization. Texture roughness features are measured by calculating parameters such as the grayscale value variation and roughness index of the soil surface image, reflecting the size and compactness of soil particles. Crack distribution features are extracted by detecting parameters such as the number, length, width, direction, and area ratio of cracks on the soil surface, reflecting the soil's wet / dry condition and structural stability.

[0087] The hierarchical distribution characteristics of a soil profile are identified by determining the boundary positions, thicknesses, and number of different layers, reflecting the vertical structure of the soil. Color transition characteristics analyze the smoothness or abruptness of color changes between adjacent layers; a smoother color transition indicates a more stable soil formation process, while a more pronounced transition may suggest changes in the depositional environment. Structural density characteristics are extracted by analyzing the number, size, and distribution of pores in each soil layer. Numerous and uniformly distributed pores indicate a loose structure with good air and water permeability; conversely, fewer pores indicate a dense structure.

[0088] The water distribution characteristics of the irrigation area are reflected by analyzing the distribution of brightness values ​​at different locations in the irrigation area image (areas with high water content are usually less bright), and calculating parameters such as the mean, variance, and spatial autocorrelation coefficient of water content. Color wetness characteristics are extracted based on changes in soil color after irrigation (usually a darker color), calculating the degree of color difference between wet and non-wet areas. Boundary sharpness characteristics are measured by analyzing the degree of blurring or sharpness of the irrigation area boundaries; the sharper the boundaries, the more precise the irrigation control.

[0089] Step S138: Normalize the visual features of the leaf region, stem region, root region, soil surface region, soil profile region, and irrigation region respectively. Based on the spatial correspondence in the time-related visual image pairs, map and associate the normalized leaf region visual features with the normalized soil surface region environmental visual features, the normalized stem region visual features with the normalized soil profile region environmental visual features, and the normalized root region visual features with the normalized irrigation region environmental visual features. Integrate all association results to generate the crop environment interaction features of the first time-related visual image pair.

[0090] The extracted visual features of various crop organ regions and environmental regions were normalized. The purpose of normalization is to eliminate differences in units and numerical ranges between different features, enabling them to be compared and correlated on the same scale. For example, for the color distribution features of leaf regions, the proportions of each color were divided by the sum of the proportions of all colors to obtain the normalized color distribution proportions; for texture feature parameters, their values ​​were mapped to the [0,1] interval through a linear transformation. Other types of features (such as morphological feature parameters and environmental region feature parameters) were processed using similar normalization methods.

[0091] After normalization, feature mapping and association are performed based on the spatial correspondence between temporally correlated visual image pairs. For example, the mapping and association between visual features of the leaf region and environmental visual features of the soil surface region is achieved by associating the color distribution features, texture arrangement features, and edge contour features of the leaf region with the color change features, texture roughness features, and crack distribution features of the corresponding soil surface region on a parameter-by-parameter basis, based on the previously established spatial correspondence of the leaf's projection area on the soil surface. Feature concatenation can be used to combine the normalized feature vector of the leaf region with the normalized feature vector of the corresponding soil surface region to form a leaf-soil surface interactive feature vector.

[0092] Similarly, the visual features of the stem region are mapped and associated with the corresponding environmental visual features of the soil profile region to generate a stem-soil profile interaction feature vector; the visual features of the root region are mapped and associated with the corresponding environmental visual features of the irrigation region to generate a root-irrigation region interaction feature vector.

[0093] Finally, the leaf-soil surface interaction feature vector, stem-soil profile interaction feature vector, and root-irrigation area interaction feature vector are integrated, for example, by splicing them together to form a higher-dimensional feature vector. This feature vector is the crop-environment interaction feature of the first time-related visual image pair, which contains the interaction information between the crop organs and the corresponding environmental regions.

[0094] Step S139: Process all time-related visual image pairs in the time-related visual image pair set in sequence to obtain the crop-environment interaction features corresponding to each time-related visual image pair.

[0095] Following the same method used to process the first temporally correlated visual image pair, feature interaction extraction is sequentially performed on all other image pairs in the set. Specifically, for each image pair, the crop visual image and the saline-alkali land environment visual image are subjected to region segmentation, feature extraction, and normalization. Then, feature mapping is performed based on spatial correspondence to generate their respective crop-environment interaction features. Through this process, each temporally correlated visual image pair is transformed into a corresponding crop-environment interaction feature.

[0096] Step S140: Based on the temporal variation pattern of the crop visual sequence, perform evolutionary trajectory modeling on the crop-environment interaction features at each time node, analyze the differences in the changes of crop-environment interaction features at adjacent time nodes, and construct the evolutionary trajectory of crop-environment interaction features.

[0097] After obtaining the crop-environment interaction characteristics at each time point, it is necessary to analyze the changing patterns of these characteristics over time and construct an evolutionary trajectory. The evolutionary trajectory can intuitively reflect the dynamic process of crop-environment interaction. By analyzing the differences in the changes of characteristics at adjacent time points, we can gain a deeper understanding of the crop's growth dynamics and environmental adaptation process in saline-alkali soil environments.

[0098] Step S141: Extract the arrangement order of all time nodes from the time-series crop visual sequence, determine the time interval between each time node, and form a time node sequence that reflects the temporal variation pattern of crop growth.

[0099] In this embodiment, the acquisition time nodes corresponding to all crop visual image groups are extracted from the previously constructed time-series crop visual sequence and arranged in chronological order to form a time node sequence. For example, the time node sequence can be represented as "T1, T2, T3, ..., Tn", where T1 is the earliest acquisition time node and Tn is the latest acquisition time node.

[0100] Then, the time interval between two adjacent time nodes is calculated. For example, the time interval between T2 and T1 is ΔT1, the time interval between T3 and T2 is ΔT2, and so on. These time intervals are determined based on the crop growth cycle and the collection plan. The time intervals may be shorter during the rapid growth period of wheat and longer during the slow growth period. The time node sequence and its corresponding time intervals together reflect the temporal variation pattern of wheat growth, providing a time axis basis for the subsequent construction of the evolutionary trajectory.

[0101] Step S142: Extract crop environment interaction features corresponding to each time node from the set of time-related visual image pairs, and arrange the extracted crop environment interaction features in the order of the time node sequence to form a crop environment interaction feature sequence.

[0102] Based on the time-node sequence, crop-environment interaction features corresponding to each time node are extracted from the set of time-related visual image pairs. For example, time node T1 corresponds to crop-environment interaction feature F1, time node T2 corresponds to crop-environment interaction feature F2, and so on. Then, these crop-environment interaction features are arranged sequentially according to the time-node sequence to form the crop-environment interaction feature sequence "F1, F2, F3, ..., Fn". This sequence is the original data sequence of crop-environment interaction features changing over time and is the core data for constructing the evolutionary trajectory.

[0103] Step S143: Select the first crop environment interaction feature in the crop environment interaction feature sequence as the initial feature point, and record the time node information and feature composition content corresponding to the initial feature point.

[0104] The first feature F1 from the crop-environment interaction feature sequence is selected as the initial feature point. Information about the time node T1 corresponding to this initial feature point is recorded, including specific time data such as year, month, day, hour, and minute. Simultaneously, the feature components of F1 are recorded in detail, specifically the parameter values ​​of the leaf-soil surface interaction feature vector, stem-soil profile interaction feature vector, and root-irrigation area interaction feature vector. These parameter values ​​reflect the interaction state between the wheat crop and the saline-alkali land environment at time T1.

[0105] Step S144: Select the second crop environment interaction feature in the crop environment interaction feature sequence as the subsequent feature point, compare the features of the subsequent feature point with the initial feature point, and analyze the differences in color features, texture features and morphological features between the two.

[0106] The second feature F2 in the crop-environment interaction feature sequence was selected as the subsequent feature point. F2 was compared with the initial feature point F1, focusing on analyzing the differences in color, texture, and morphological features. Taking color features as an example, the changes in leaf area color distribution characteristics in F1 and F2 were compared, such as the increase or decrease in the proportion of green and the change in the proportion of yellow; simultaneously, the changes in soil surface area color characteristics were compared, such as the changes in color depth in salinized areas.

[0107] Regarding texture features, the changes in the regularity of leaf surface texture arrangement, the increase or decrease in stem surface texture density, and the changes in soil surface texture roughness were compared. Regarding morphological features, the changes in leaf edge contour shape (such as changes in aspect ratio), the changes in stem morphology and structure thickness and curvature, and the changes in the number and length of root branches were compared. Through these comparisons, a comprehensive analysis of the changes in various features from F1 to F2 was conducted.

[0108] Step S145: For color features, calculate the change in the proportion of color distribution in the corresponding areas of the initial feature points and subsequent feature points to determine the direction and magnitude of color feature changes; for texture features, analyze the changes in the texture arrangement pattern or density in the corresponding areas of the initial feature points and subsequent feature points to determine the trend of texture feature changes; for morphological features, compare the changes in the contour shape or structure of the corresponding areas of the initial feature points and subsequent feature points to determine the morphological feature change pattern.

[0109] To calculate the change in color characteristics, taking the proportion of green in leaf areas as an example, let C1 be the proportion of green in leaves in F1 and C2 be the proportion of green in leaves in F2. Then, the change in color characteristics ΔC = C2 - C1. If ΔC is positive, it indicates an increase in the proportion of green, and the direction of color characteristic change is positive; if ΔC is negative, the direction of change is negative. The magnitude of the change is reflected by the absolute value of ΔC; the larger the absolute value, the greater the magnitude of the change. Similarly, the changes in color characteristics in stem, root, and various environmental areas are calculated to determine their respective directions and magnitudes of change.

[0110] Analysis of the changing trends of texture features, such as leaf texture arrangement features, involves comparing the changes in parameters (e.g., entropy values) related to the texture arrangement regularity in F1 and F2. If the entropy value decreases, it indicates that the texture arrangement is changing from random to regular, with a trend towards regularization; conversely, it indicates a trend towards randomization. For stem texture density features, if the texture density parameter value increases, it indicates that the texture density is increasing, with a trend towards compaction; conversely, it indicates a trend towards sparsity. By analyzing the above parameters, the changing trends of various texture features can be determined.

[0111] The patterns of change in morphological characteristics are determined. For example, changes in leaf edge contours are analyzed by comparing parameters such as area, perimeter, and concavity / convexity of the leaf contours in F1 and F2. If the area and perimeter increase and the contour becomes fuller, it indicates that the leaf is growing and expanding, and the pattern is an expansion type. If the contour shows defects or deformation, the pattern may be an abnormal type. The patterns of change in stem morphological and structural characteristics are also determined. For example, an increase in stem diameter and a decrease in bending angle indicate robust stem growth, and the pattern is a robust type. Conversely, it may indicate a weak or lodging-prone type. Through the above analysis, the patterns of change for each morphological characteristic are determined.

[0112] Step S146: Based on the differences in color features, texture features, and morphological features, calculate the feature change distance between the initial feature point and the subsequent feature point. This feature change distance reflects the degree of change in the crop environment interaction features at two time points.

[0113] For example, step S1461: extract the color feature part from the crop environment interaction features of the initial feature points, which includes the color-related features of the leaf area, stem area and root area; at the same time, extract the corresponding color feature part from the crop environment interaction features of subsequent feature points.

[0114] In this embodiment, the color features of the initial feature point F1 and the subsequent feature point F2 are extracted respectively. For F1, the color-related features of the leaf region include parameters such as the proportion of green and yellow distributions; the stem region includes statistics such as the standard deviation and variance of color uniformity; and the root region includes the average gray value of color depth. Similarly, the color feature portion of F2 contains parameters of the same type for the corresponding region. This extraction ensures that subsequent comparisons are made of changes in the same type of color feature parameters.

[0115] Step S1462: Calculate the difference in color features of the leaf region. The pixel ratio of each color interval in the color distribution features of the initial feature point leaf region is compared with the pixel ratio of the corresponding color interval in the color distribution features of the subsequent feature point leaf region. The absolute difference of the pixel ratio of each color interval is calculated, and the absolute differences of all color intervals are summed to obtain the leaf color feature difference value.

[0116] Taking the leaf region as an example, in F1, the pixel ratio of the green area is C1g, the yellow area is C1y, and the other color areas are C1o; in F2, the corresponding pixel ratios are C2g, C2y, and C2o. The absolute differences between each interval are calculated as |C1g-C2g|, |C1y-C2y|, and |C1o-C2o|. These differences are then summed to obtain the leaf color feature difference value ΔC_leaf = |C1g-C2g| + |C1y-C2y| + |C1o-C2o|. This value reflects the degree of change in the overall color distribution of the leaf.

[0117] Step S1463: Using the same method, calculate and sum the absolute differences of each color-related parameter in the stem region color features to obtain the stem color feature difference value; calculate and sum the absolute differences of each color-related parameter in the root region color features to obtain the root color feature difference value; add the leaf color feature difference value, stem color feature difference value, and root color feature difference value to obtain the total color feature difference value between the initial feature point and subsequent feature points.

[0118] For the stem region, the standard deviation of color uniformity is S1s for F1 and S2s for F2. The difference in stem color characteristics is ΔC_stem = |S1s - S2s|. If other color-related parameters exist (such as mean saturation), the absolute differences of all parameters are calculated and summed. The root region is similar; the average grayscale value is G1m for F1 and G2m for F2. The difference in root color characteristics is ΔC_root = |G1m - G2m|. The total difference in color characteristics is ΔC_total = ΔC_leaf + ΔC_stem + ΔC_root.

[0119] Step S1464: Extract the texture feature part from the crop environment interaction features of the initial feature point and subsequent feature points respectively. The texture feature part includes the texture-related features of the leaf region, stem region and root region.

[0120] Texture features are extracted from F1 and F2. For leaf regions, parameters include texture entropy and contrast; for stem regions, parameters include texture density and energy; and for root regions, parameters include texture orientation consistency and correlation. These parameters were already obtained during the feature extraction stage using algorithms such as gray-level co-occurrence matrix (GLCM), and are directly extracted here for difference calculation.

[0121] Step S1465: Perform difference calculation on the texture features of the leaf region. Calculate the absolute difference between each texture parameter in the texture arrangement features of the initial feature point leaf region and the corresponding texture parameter in the texture arrangement features of the subsequent feature point leaf region. Sum the absolute differences of all texture parameters to obtain the leaf texture feature difference value.

[0122] The entropy of leaf texture for F1 is E1e, and the contrast is E1c; the corresponding parameters for F2 are E2e and E2c. The difference value of leaf texture features ΔT_leaf = |E1e-E2e| + |E1c-E2c|. If there are more texture parameters (such as correlation and energy), the corresponding absolute differences are accumulated to comprehensively reflect the changes in the texture arrangement pattern.

[0123] Step S1466: Using the same method, calculate and sum the absolute differences of each texture parameter in the texture features of the stem region to obtain the stem texture feature difference value; calculate and sum the absolute differences of each texture parameter in the texture features of the root region to obtain the root texture feature difference value; add the leaf texture feature difference value, stem texture feature difference value and root texture feature difference value to obtain the total texture feature difference value between the initial feature point and the subsequent feature points.

[0124] The stem region has a texture density of D1d and energy of D1e in region F1; and D2d and D2e in region F2. The stem texture difference value ΔT_stem = |D1d - D2d| + |D1e - D2e|. The root region has a texture orientation consistency of O1o and correlation of O1c in region F1; and O2o and O2c in region F2. The root texture difference value ΔT_root = |O1o - O2o| + |O1c - O2c|. The total texture feature difference value ΔT_total = ΔT_leaf + ΔT_stem + ΔT_root.

[0125] Step S1467: Extract morphological features from the crop environment interaction features of the initial feature points and subsequent feature points respectively. The morphological features include morphological features of the leaf region, stem region and root region.

[0126] The morphological features of F1 and F2 are extracted. For the leaf region, parameters include outline area and perimeter; for the stem region, parameters include diameter and bending angle; and for the root region, parameters include number of branches and distribution area. These parameters were determined during edge detection and morphological analysis and are directly used in subsequent calculations.

[0127] Step S1468: Perform difference calculation on the morphological features of the leaf region. Calculate the absolute difference between each shape parameter in the edge contour features of the initial feature point of the leaf region and the corresponding shape parameter in the edge contour features of the subsequent feature point of the leaf region. Sum the absolute differences of all shape parameters to obtain the leaf morphological feature difference value.

[0128] F1 has a leaf profile area of ​​A1a and a perimeter of A1p; F2 has A2a and A2p. The leaf morphology difference value ΔM_leaf = |A1a - A2a| + |A1p - A2p|. If other shape parameters such as concavity or convexity exist, they are also included in the summation to fully reflect the changes in profile shape.

[0129] Step S1469: Using the same method, calculate and sum the absolute differences of each morphological parameter in the stem region to obtain the stem morphological feature difference value; calculate and sum the absolute differences of each morphological parameter in the root region to obtain the root morphological feature difference value; for the leaf region, stem region, and root region respectively, normalize the color feature difference value, texture feature difference value, and morphological feature difference value corresponding to each region; for each region, set the color feature weight, texture feature weight, and morphological feature weight for that region; multiply the normalized color feature difference value by the color feature weight. The normalized texture feature difference value is multiplied by the texture feature weight, and the normalized morphological feature difference value is multiplied by the morphological feature weight to obtain the weighted feature difference value for each region. The weighted feature difference values ​​of the leaf region, stem region, and root region are added together to obtain the feature change distance between the initial feature point and the subsequent feature points. The magnitude of this feature change distance directly reflects the overall degree of change of the crop environment interaction features at two time points. When the feature change distance meets the preset significant change threshold, the change is judged to have reached the significant standard; when the feature change distance meets the preset gradual change threshold, the change is judged to have reached the gradual standard.

[0130] The stem region F1 has a diameter of S1d and a bending angle of S1a; F2 has S2d and S2a. The stem morphology difference value ΔM_stem = |S1d - S2d| + |S1a - S2a|. The root region F1 has R1n branches and a distribution area of ​​R1a; F2 has R2n and R2a. The root morphology difference value ΔM_root = |R1n - R2n| + |R1a - R2a|.

[0131] The differences in the three types of features in each region are normalized. For example, the color difference ΔC_leaf in the leaf region is normalized to the [0,1] interval as ΔC_leaf_norm, the texture difference ΔT_leaf is normalized to ΔT_leaf_norm, and the morphological difference ΔM_leaf is normalized to ΔM_leaf_norm. Leaf region weights are set as follows: color weight Wlc = 0.4, texture weight Wlt = 0.3, and morphological weight Wlm = 0.3. Therefore, the weighted difference value of the leaf ΔL = ΔC_leaf_norm × Wlc + ΔT_leaf_norm × Wlt + ΔM_leaf_norm × Wlm.

[0132] Similarly, for the stem region, the weights are Wsc=0.3, Wst=0.3, Wsm=0.4, and the weighted difference value ΔS=ΔC_stem_norm×Wsc+ΔT_stem_norm×Wst+ΔM_stem_norm×Wsm. For the root region, the weights are Wrc=0.2, Wrt=0.5, Wrm=0.3, and the weighted difference value ΔR=ΔC_root_norm×Wrc+ΔT_root_norm×Wrt+ΔM_root_norm×Wrm.

[0133] The feature change distance D = ΔL + ΔS + ΔR. Preset thresholds for significant change (D_sig) and gradual change (D_gen) are used. A change is considered significant when D ≥ D_sig, gradual when D ≤ D_gen, and moderate when D falls between the two. This weighted calculation comprehensively considers the contribution of different features from each region to the overall change, achieving precise quantification of the degree of change in crop-environment interaction characteristics.

[0134] Step S147: Starting from the initial feature point and taking subsequent feature points as the next node, draw a feature change line segment between two points based on the feature change distance and the corresponding time interval. The length of the feature change line segment corresponds to the feature change distance, and the extension direction of the feature change line segment corresponds to the feature change trend.

[0135] Using the initial feature point F1 as the origin, in a multi-dimensional feature space (each dimension corresponds to a feature parameter), the length of the line segment is determined based on the distance of feature change, and the extension direction of the line segment is determined based on the trend of feature change (such as the direction of change of each feature parameter). The feature change line segment from F1 to F2 is then drawn. Simultaneously, the corresponding time interval ΔT1 is considered, incorporating the time factor into the drawing of the line segment; for example, the slope of the line segment can reflect the rate of feature change per unit time. In this way, the feature change between F1 and F2 is visually represented in the form of line segments.

[0136] Step S148: Take the second crop environment interaction feature as the new initial feature point, select the third crop environment interaction feature as the new subsequent feature point, and repeat the above steps of feature comparison, change difference analysis, feature change distance calculation, and feature change line segment drawing; process all crop environment interaction features in the crop environment interaction feature sequence in sequence, connect all feature change line segments in sequence to form an evolution trajectory that reflects the changes of crop environment interaction features over time. Each node in this evolution trajectory contains information about the corresponding time node, the content of the crop environment interaction feature, and the change relationship with the previous node.

[0137] Using F2 as the new initial feature point and F3 as the new subsequent feature point, repeat steps S144 to S147, namely, performing feature comparison, variation difference analysis, feature change distance calculation, and drawing the feature change line segment from F2 to F3. Following the same method, process all feature points in the crop-environment interaction feature sequence sequentially until the last feature point Fn is reached.

[0138] Connecting all the drawn feature change lines sequentially forms the evolutionary trajectory of crop-environment interaction characteristics. In this trajectory, each node (i.e., each feature point) contains information about the corresponding time point (e.g., Tn), the content of the crop-environment interaction characteristics (e.g., the feature parameter values ​​of Fn), and the relationship with the previous node (e.g., feature change distance, change direction, change trend, etc.). Through this evolutionary trajectory, the dynamic change process of wheat crop-environment interaction characteristics over time can be observed.

[0139] Step S150: Generate crop phenotypic recognition results based on the evolutionary trajectory of crop-environment interaction features, and output a phenotypic recognition report containing phenotypic features and information related to the saline-alkali land environment based on the crop phenotypic recognition results. The crop phenotypic recognition results include crop growth status features, morphological structure features and physiological response features to the saline-alkali land environment.

[0140] After the evolutionary trajectory of crop-environment interaction features is constructed, it is necessary to identify crop phenotypes based on this trajectory, extract key phenotypic features such as crop growth status, morphological structure and physiological response, and generate a report containing information on the association between these features and the saline-alkali land environment.

[0141] Step S151: Node-splitting is performed on the evolution trajectory of crop-environment interaction features, and each feature node in the evolution trajectory is extracted. Each feature node contains time node information, crop-environment interaction feature content, and change relationship with the previous node.

[0142] In this embodiment, the evolutionary trajectory of the wheat crop-environment interaction features constructed above is decomposed into nodes. The evolutionary trajectory is formed by connecting multiple feature nodes through feature change segments, and each feature node corresponds to a time node. Through the decomposition operation, the evolutionary trajectory is decomposed into a series of independent feature nodes, such as N1 (corresponding to T1), N2 (corresponding to T2), ..., Nn (corresponding to Tn). Each feature node Nk (k=1,2,...,n) contains information about the corresponding time node Tk, the content of the crop-environment interaction feature Fk, and the relationship of change with the previous feature node Nk-1 (such as feature change distance, change direction, etc.).

[0143] Step S152: For the crop environment interaction feature content of each feature node, separate the crop organ visual feature part, which includes the visual features of the leaf area, stem area and root area.

[0144] For each feature node Nk, the visual features of the crop organs are extracted from its crop-environment interaction feature content Fk. Since Fk is a mapping and association between the visual features of crop organs and the visual features of the environment, the visual features of the leaf region, stem region, and root region can be separated from Fk through reverse extraction. For example, the visual features of the leaf region include parameters such as color distribution features, texture arrangement features, and edge contour features; the visual features of the stem region include parameters such as color uniformity features, texture density features, and morphological structure features; and the visual features of the root region include parameters such as color depth features, texture direction features, and branch distribution features. The extracted visual features of the crop organs are the direct basis for crop phenotypic recognition.

[0145] Step S153: Analyze the changing patterns of visual features of the leaf region over time. By comparing the color distribution features, texture arrangement features, and edge contour features of the leaf region at different time points, determine the growth rate and stability of the growth state of the leaf. Statistically calculate the green proportion in the color distribution features. When the green proportion meets the preset photosynthetic threshold, the photosynthetic capacity of the leaf is considered to have reached the standard. Record the transformation process of the texture arrangement features from disordered to regular, which corresponds to the transition of the leaf from immature to mature. Calculate the expansion value of the contour range in the edge contour features. The ratio of this expansion value to the time interval corresponds to the quantitative result of the leaf growth rate. Integrate all the analytical information to form the leaf growth state features.

[0146] The visual characteristics of the isolated leaf regions were continuously tracked and analyzed according to a time sequence (T1 to Tn). The color distribution characteristics of the leaf regions at different time points were compared, with a focus on changes in the green proportion. A preset photosynthetic threshold was established; when the green proportion of the leaves at a certain time point reached or exceeded this threshold, the photosynthetic capacity of the leaves at that time point was considered to have reached a standard level. For example, during the wheat jointing stage, a preset photosynthetic threshold was set; when the green proportion of the leaves at time point T5 reached this value, it indicated that the photosynthetic capacity of the leaves was relatively strong at that time.

[0147] The study recorded the changes in texture arrangement characteristics, from the chaotic texture of wheat leaves during their early stages (e.g., time nodes T1 and T2), to the gradual growth (e.g., time nodes T3 and T4) where the texture becomes regular, until the texture becomes regular and stable during maturity (e.g., time node Tn). This transformation from chaotic to regular clearly reflects the transition of leaves from their early to mature stages.

[0148] Calculate the expanded value of the contour range in the edge contour feature, that is, the difference in the leaf contour area at different time points. For example, if the leaf contour area at time point T2 is S2 and the leaf contour area at time point T3 is S3, then the expanded value ΔS = S3 - S2. Compare ΔS with the time interval ΔT from T2 to T3 to obtain the leaf growth rate v = ΔS / ΔT, which quantifies how fast the leaf grows during that time period.

[0149] By integrating the above analytical information on leaf photosynthetic capacity, stage transition, and growth rate, a leaf growth status characteristic is formed, which comprehensively reflects the growth status of wheat leaves throughout the entire growth cycle.

[0150] Step S154: Analyze the changes in visual characteristics of the stem region over time, and compare the color uniformity, texture density, and morphological structure characteristics of the stem region at different time points; calculate the numerical changes in color uniformity characteristics, and determine that the stem nutrient absorption has reached a balanced standard when the numerical changes meet the preset nutrient threshold; statistically analyze the numerical increments of texture density characteristics, and determine that the stem lignification degree has reached the preset stage when the numerical increments meet the preset lignification threshold; measure the numerical changes in stem diameter and the angle of stem deviation from the vertical direction in the morphological structure characteristics, with the diameter value corresponding to the growth thickness and the angle value corresponding to the degree of uprightness; integrate all analytical information to form stem growth status characteristics.

[0151] For the visual characteristics of the stem region, the analysis is also performed according to the time node sequence. The numerical change of the color uniformity characteristic is calculated. Let the stem color uniformity at time node Tk be U1, and at time node Tk+1 be U2, then the numerical change ΔU = U2 - U1. A preset nutrient threshold is established. When the trend and magnitude of ΔU's change meet this threshold range, it is determined that the stem's nutrient absorption has reached a balanced standard. For example, during the wheat tillering stage, the stem color uniformity gradually increases. When it reaches the preset nutrient threshold, it indicates that the stem's absorption of nutrients such as nitrogen, phosphorus, and potassium is relatively balanced.

[0152] The numerical increment of the statistical texture density feature is calculated. Let the stem texture density at time node Tk be D1 and the stem texture density at time node Tk+m be D2. Then the numerical increment ΔD = D2 - D1. A lignification threshold is preset. When ΔD reaches this threshold, the degree of lignification of the stem is determined to have reached a preset stage (such as the lignification stage at the jointing stage).

[0153] The change in stem diameter is measured as a morphological feature. For example, the stem diameter at time point T1 is d1, and at time point Tn it is dn. The change in diameter Δd = dn - d1 reflects the thickness of the stem growth. Simultaneously, the angle of the stem's deviation from the vertical direction is measured. A smaller angle indicates a higher degree of stem uprightness and stronger resistance to lodging. Information on stem nutrient absorption balance, lignification degree, growth thickness, and uprightness is integrated to form a characteristic profile of stem growth status.

[0154] Step S155: Analyze the changes in visual characteristics of the root system region over time, and compare the color depth, texture direction, and branch distribution characteristics of the root system region at different time points; count the uniformity value of the color depth characteristic, and determine that the root system vitality has reached the standard when the uniformity value meets the preset vitality threshold; record the direction data of the texture direction characteristic extending towards the water-sufficient area, which corresponds to the root hydrotropism result; measure the increment of the number of branches and the expansion area of ​​the distribution range in the branch distribution characteristic, with the increment value corresponding to the branch growth result and the expansion area corresponding to the expansion capacity result; integrate all the analysis information to form the root growth status characteristics; integrate the leaf growth status characteristics, stem growth status characteristics, and root growth status characteristics to obtain the crop growth status characteristics.

[0155] Analyze the changes in visual characteristics of the root system region over time. Statistically calculate the uniformity of color intensity, which reflects the evenness of the root color distribution. A pre-defined vitality threshold is established; when the uniformity of root color intensity reaches this threshold at a certain time point, the root vitality is considered to have reached the standard, indicating vigorous root physiological activity.

[0156] Record changes in root texture characteristics, focusing on data showing the direction of root texture extension towards areas with sufficient water. For example, in irrigated areas with sufficient water, the root texture is observed to gradually extend towards the irrigated area; this directional data represents the root's hydrotropism, reflecting its adaptability to the water environment.

[0157] The measurement of branch distribution characteristics includes the increment in the number of branches and the expansion area of ​​the distribution range. The increment in the number of branches is the difference in the total number of root branches at different time points, reflecting the growth status of root branches; the expansion area of ​​the distribution range is the difference in the area of ​​the root distribution region at different time points, reflecting the root system's ability to expand in the soil. Information such as root activity, hydrotropism, branch growth, and expansion ability are integrated to form the root growth status characteristics.

[0158] Finally, the characteristics of leaf growth, stem growth, and root growth are integrated to form a comprehensive crop growth status characteristic that reflects the growth status of wheat.

[0159] Step S156: Extract the morphologically related parts of the visual features of crop organs in each feature node. The edge contour features of the leaf region, the morphological structure features of the stem region, and the branch distribution features of the root region are all morphologically related parts.

[0160] From the visual features of crop organs at each feature node, the morphological and structurally relevant parts are extracted. For the leaf region, the edge contour features directly reflect the shape of the leaf and belong to the morphologically relevant part; for the stem region, morphological and structural features (such as thickness and curvature) describe the shape of the stem and belong to the morphologically relevant part; for the root region, branch distribution features (such as the number, length, and distribution range of branches) reflect the morphological and structural features of the root system and belong to the morphologically relevant part. Extracting these morphologically relevant parts prepares the basis for analyzing the morphological and structural features of the crop.

[0161] Step S157: Compare the edge contour features of the leaf region at different time points, record the stage data of the leaf shape changing from slender to wide and round, and this stage data corresponds to the growth stage result; calculate the growth value of the contour area, and this growth value corresponds to the size growth result; count the number of continuous unbroken segments of the contour line, and when the number of continuous unbroken segments meets the preset integrity threshold, the contour integrity is judged to meet the standard; form the leaf morphological structure features based on all the analysis information; compare the morphological structure features of the stem region at different time points, measure the numerical change of the stem diameter, and this numerical change corresponds to the thickness change result; calculate the fluctuation range of the angle between the stem axis and the vertical line. When the fluctuation range meets the preset stability threshold, the overall morphological stability is determined to have reached the standard; stem morphological structure characteristics are formed based on all analytical information; the branch distribution characteristics of the root system region at different time points are compared, and the numerical changes in the number of branches are statistically analyzed, with the numerical changes corresponding to the branch number changes; the numerical changes in branch length are measured, with the numerical changes corresponding to the branch length changes; the area changes of the root system coverage area are calculated, with the area changes corresponding to the overall distribution range results; root morphological structure characteristics are formed based on all analytical information; and the leaf morphological structure characteristics, stem morphological structure characteristics, and root morphological structure characteristics are integrated to obtain the crop morphological structure characteristics.

[0162] Compare the edge contour features of leaf regions at different time points to observe changes in leaf shape. Wheat leaves typically change shape from slender to broad and rounded from the seedling stage to maturity. Record the time range in which this transformation occurs, i.e., stage data, which corresponds to the leaf's growth stage, such as the transition from seedling leaves to jointing leaves.

[0163] The growth rate of leaf outline area is calculated, which is the sum of the differences in leaf outline area at different time points. This sum reflects the increase in leaf size as it grows. The number of continuous, unbroken segments of the outline is counted. A threshold for completeness is preset. When the number of continuous, unbroken segments reaches this threshold, it indicates that the leaf outline is of good integrity, without obvious damage or outline loss caused by disease, and the outline integrity is judged to meet the standard. Information such as leaf growth stage, size increase, and outline integrity is integrated to form the leaf morphological structure characteristics.

[0164] By comparing the morphological and structural characteristics of the stem region at different time points, the numerical change in stem diameter is measured, i.e., the total change in stem diameter from the initial time point to the current time point, which corresponds to the change in stem thickness. The fluctuation range of the angle between the stem axis and the vertical line is calculated, i.e., the difference between the maximum and minimum values ​​of the angle at different time points. A stability threshold is preset; when the fluctuation range is less than or equal to the threshold, it indicates that the overall morphological stability of the stem is good, the lodging resistance is strong, and the overall morphological stability meets the standard. Information such as stem thickness variation and morphological stability is integrated to form the stem morphological and structural characteristics.

[0165] By comparing the branch distribution characteristics of the root system region at different time points, the numerical changes in the number of branches (i.e., the total increase in the total number of branches) are statistically analyzed, corresponding to the change in the number of branches; the numerical changes in the length of branches (i.e., the increase in the total length of all branches) are measured, corresponding to the change in the length of branches; and the changes in the area covered by the root system (i.e., the total increase in the root distribution area) are calculated, corresponding to the overall distribution range. The information on changes in the number, length, and distribution range of root branches is integrated to form the root system morphological structure characteristics. Finally, the morphological structure characteristics of leaves, stems, and roots are integrated to obtain the crop morphological structure characteristics.

[0166] Step S158: Extract the correlation and change of crop organ visual features and environmental visual features in each feature node, and analyze the response of crop organ visual features to changes in environmental visual features; when the water distribution characteristics of the soil surface area change, record the time interval of leaf color changing from dark green to light green, and compare the time interval with a preset water response threshold. When the time interval is less than the preset water response threshold, it is determined that the leaf response to soil water changes has reached a preset sensitivity standard; when the structural density characteristics of the soil profile area change, count the number of root branches penetrating the dense soil area, and compare the number with a preset structural response threshold. When the number is greater than the preset structural response threshold, it is determined that the root system response to soil structure changes has reached a preset capacity standard; when the water distribution characteristics of the irrigation area change, measure the speed at which the root color changes from light to dark, and compare the speed with a preset absorption response threshold. When the speed is greater than the preset absorption response threshold, it is determined that the root system absorption response to irrigation water has reached a preset efficiency standard.

[0167] From the crop-environment interaction features of each feature node, the correlation and change between crop organ visual features and environmental visual features are extracted. This involves analyzing how crop organ visual features change in response to changes in environmental visual features. For example, when the moisture distribution characteristics of the soil surface area change (e.g., moisture content decreases), the change in leaf color is closely monitored, and the time interval Δt1 during which the leaf changes from dark green to light green is recorded. A preset moisture response threshold t_water is established. When Δt1 is less than t_water, it indicates that the leaf's perception and response speed to changes in soil moisture is fast, and the leaf's response to changes in soil moisture meets the preset sensitivity standard.

[0168] When the structural density characteristics of a soil profile change (e.g., a certain layer of structure becomes denser), the number N of root branches penetrating that denser soil region is counted. A pre-set structural response threshold Nk is used. When N is greater than Nk, it indicates that the root system has a strong penetrating ability and can adapt to changes in soil structure. The root system's response to changes in soil structure is then judged to have reached the pre-set capability standard.

[0169] When the water distribution characteristics of the irrigated area change (such as an increase in water content after irrigation), the rate at which the root color changes from light to dark is measured, i.e., the change in the uniformity of root color depth per unit time. A preset absorption response threshold, vabsorb, is set. When v is greater than vabsorb, it indicates that the roots can quickly absorb water and undergo a color change (usually, a darker color indicates sufficient water), and the root system's absorption response to irrigation water is judged to have reached the preset efficiency standard.

[0170] Step S159: Integrate the visual feature response information of crop organs to different environmental visual features, and count the length of the cycle in which the crop maintains normal growth under soil moisture fluctuation scenarios. When the length of this cycle meets the preset adaptation threshold, the crop's adaptability to water is determined to meet the standard. Count the range of the crop's root system expansion under dense soil structure scenarios. When the range of this range meets the preset tolerance threshold, the crop's tolerance to soil structure is determined to meet the standard. Based on all the analyzed information, form the physiological response characteristics of the crop to the saline-alkali environment. Integrate the crop growth status characteristics, crop morphological structure characteristics, and the physiological response characteristics of the crop to the saline-alkali environment to obtain the crop phenotypic recognition results.

[0171] This study integrates information on the responses of crop organs such as leaves and roots to environmental visual characteristics such as soil moisture, soil structure, and irrigation water. It statistically analyzes the length of time (Tfit) during which wheat can maintain normal growth (i.e., no significant abnormalities in growth status and morphological structure) under scenarios with multiple fluctuations in soil moisture (e.g., alternating drought and wet conditions). A preset adaptation threshold (Tthreshold) is used; when Tfit reaches or exceeds Tthreshold, the crop's adaptability to water is considered to have reached the standard.

[0172] The numerical value R (such as root coverage area or maximum depth) of wheat root system expansion under dense soil conditions is statistically analyzed. A tolerance threshold R is preset. When R reaches or exceeds the R threshold, the crop's tolerance to soil structure is determined to meet the standard.

[0173] Based on the above information on crop responses to environmental factors such as water and soil structure, as well as the possible responses to other environmental factors such as salinity and alkali concentration, physiological response characteristics of crops to saline-alkali environments are formed. Finally, the crop growth status characteristics, crop morphological structure characteristics, and physiological response characteristics of crops to saline-alkali environments are integrated to obtain complete crop phenotypic identification results.

[0174] Step S1510: Organize the crop growth status features, crop morphological structure features, and crop physiological response features to saline-alkali land environment from the crop phenotypic recognition results into independent feature modules. Each feature module includes a feature name, specific feature content, and the time node range corresponding to the feature.

[0175] The obtained crop phenotypic identification results were organized, and crop growth status characteristics, crop morphological structure characteristics, and crop physiological response characteristics to saline-alkali soil environment were each treated as an independent feature module. Each feature module contains detailed content: the feature name clearly indicates the type of feature, such as "leaf growth rate", "stem lodging resistance", "root water absorption efficiency", etc.; the feature content is a detailed description and quantitative data of the feature, such as a certain range of leaf growth rate and the level of stem lodging resistance; the corresponding time node range indicates which time nodes or growth stages the feature is mainly reflected in, such as "jointing stage to heading stage".

[0176] Step S1511: Extract the environmental visual images and environmental visual features of the saline-alkali land corresponding to each time node from the set of time-related visual image pairs. Classify and organize the environmental visual features according to soil surface area, soil profile area and irrigation area to form an environmental feature module. The environmental feature module includes environmental area type, environmental visual feature content and corresponding time node information.

[0177] From a set of time-related visual image pairs, the environmental visual images of saline-alkali land corresponding to each time node, along with previously extracted environmental visual features, are extracted. These environmental visual features are then categorized and organized according to soil surface areas, soil profile areas, and irrigated areas. For example, the environmental visual features of the soil surface area include color variation characteristics, texture roughness characteristics, and crack distribution characteristics at each time node; the soil profile area includes layer distribution characteristics, color transition characteristics, and structural density characteristics; and the irrigated area includes water distribution characteristics, color wetness characteristics, and boundary clarity characteristics. These categorized and organized environmental visual features form an environmental feature module, which includes the environmental area type (e.g., soil surface, soil profile, irrigated area), the environmental visual feature content of each area, and the corresponding time node information.

[0178] Step S1512: Establish an association index between phenotypic features and environmental features. Based on the correspondence of time nodes, associate each sub-feature in the crop growth state features with the environmental visual features at the same time node, associate each sub-feature in the crop morphological structure features with the environmental visual features at the same time node, and associate each sub-feature in the crop physiological response features with the environmental visual features that trigger the corresponding crop physiological response. Each association entry includes the phenotypic feature name, environmental feature name, association time node, and description of association changes.

[0179] Based on the correspondence between time points, an association index is established between phenotypic features and environmental features. For each sub-feature in the crop growth status features (such as the green proportion of leaves), find the environmental visual features (such as soil surface moisture distribution features) at the same time point and associate them. For example, at time point T5, the green proportion of leaves is high, and at the same time point, the soil surface moisture distribution features show sufficient moisture; these two features are then associated.

[0180] Similarly, sub-features in crop morphological structure characteristics (such as stem diameter) were correlated with environmental visual characteristics (such as soil profile density) at the same time point; sub-features in crop physiological response characteristics (such as root branch penetration) were correlated with environmental visual characteristics that triggered the response (such as changes in soil profile density). Each correlation entry recorded the phenotypic feature name, environmental feature name, correlation time point, and correlation change description (e.g., "At time point T5, sufficient soil surface moisture led to an increase in the green proportion of leaves").

[0181] Step S1513: Construct the structural framework of the phenotypic identification report, which includes the report title, report generation time, data source description, overview of phenotypic features, overview of environmental features, phenotypic-environment association analysis, and conclusions and recommendations.

[0182] Design a structural framework for the phenotypic identification report to present the phenotypic identification results and related information clearly and systematically. The report title should clearly state the topic of the report, such as "Phenotypic Identification Report of Wheat in Saline-Alkali Environment"; the report generation time should record the specific date the report was completed; and the data source description should detail the data acquisition process used for phenotypic identification, the model of the visual acquisition equipment, the acquisition parameters, and other information to ensure data traceability.

[0183] The phenotypic overview section displays the content of the crop growth status characteristics, morphological structure characteristics, and physiological response characteristics modules; the environmental characteristics overview section displays the content of the environmental characteristics module; the phenotypic-environment association analysis section presents the established association index entries; and the conclusions and recommendations section draws conclusions and makes recommendations based on the analysis results.

[0184] Step S1514: In the phenotypic feature overview section, the contents of the crop growth status feature module, the crop morphological structure feature module, and the crop physiological response feature module are presented in sequence. Each module displays feature changes in chronological order and is accompanied by a corresponding crop visual image thumbnail.

[0185] The phenotypic overview section first presents the crop growth status feature module, displaying the changes in leaf, stem, and root growth status features in chronological order (T1 to Tn), such as the increasing or decreasing trend of leaf green proportion, the change curve of stem color uniformity, and the fluctuation of root vitality. Each feature change point is accompanied by a corresponding crop visual image thumbnail, intuitively showing the actual growth status of the crop at that time point.

[0186] Next, the crop morphological and structural features module is presented, which also displays changes in morphological and structural features such as leaf shape, stem thickness, and root distribution in chronological order, accompanied by corresponding crop visual image thumbnails. Finally, the crop physiological response features module is presented, showing the changes in crop response characteristics to various factors in the saline-alkali land environment, such as the sensitivity to changes in water and the degree of tolerance to soil structure, accompanied by relevant environmental visual image thumbnails (indirectly reflecting environmental changes).

[0187] Step S1515: In the environmental features overview section, the contents of the environmental features module are presented, showing the changes in environmental visual features in the order of soil surface area, soil profile area and irrigation area, and is accompanied by corresponding thumbnail images of saline-alkali land environment.

[0188] The environmental features overview section displays changes in environmental visual features in the order of soil surface area, soil profile area, and irrigation area. For the soil surface area, it displays color change features (such as the expansion or contraction of salinization areas), texture roughness features (such as changes in the degree of cracking), and crack distribution features (such as changes in the number and size of cracks) in chronological order, accompanied by corresponding thumbnails of the soil surface visual images.

[0189] Soil profiles are displayed according to time points, showing characteristics such as layer distribution (e.g., whether new soil layers have formed), color transition characteristics (e.g., changes in color differences between layers), and structural density characteristics (e.g., changes in the number of pores), accompanied by thumbnails of visual images of the soil profiles. Irrigation areas are displayed, showing characteristics such as water distribution (e.g., changes in the irrigated area), color wetness characteristics (e.g., expansion or contraction of wet areas), and boundary clarity characteristics (e.g., changes in the irrigation boundary), accompanied by thumbnails of visual images of the irrigation areas.

[0190] Step S1516: In the phenotypic and environmental association analysis section, all association index entries are displayed in chronological order of association time nodes. Each entry details how phenotypic characteristics change with environmental characteristics, and also labels the corresponding visual image source.

[0191] In the phenotypic-environment association analysis section, all established association index entries are presented in chronological order of association time points (from T1 to Tn). Each entry details how phenotypic characteristics change with environmental characteristics. For example, "At time point T3, the soil profile density characteristic shows that a certain layer of structure becomes denser (corresponding to soil profile visual image P3), resulting in a decrease in the number of root branches penetrating this area (corresponding to root visual image R3), and a decrease in the expansion capacity result in the root growth state characteristics." Each entry clearly indicates the source of the corresponding crop visual image and saline-alkali land environment visual image (such as image number or collection time point) to facilitate readers' access to the original images.

[0192] Step S1517: In the conclusion and recommendations section, based on the results of the phenotypic-environment association analysis, the advantages and disadvantages of crop growth in the current saline-alkali environment are summarized, and suggestions are made on soil moisture adjustment and soil structure improvement. Finally, a phenotypic identification report containing phenotypic characteristics and information on the association between saline-alkali environment is generated.

[0193] In the conclusions and recommendations section, based on the results of the phenotypic-environment association analysis, the advantages of wheat growth in the current saline-alkali environment are first summarized, such as the sensitive response of the root system to water changes and the strong photosynthetic capacity of the leaves when water is sufficient. At the same time, the shortcomings are pointed out, such as the limited root expansion capacity in areas with dense soil structure and the potential risk of lodging of the stems in the later stages.

[0194] Then, recommendations are made regarding soil moisture adjustment, such as optimizing irrigation frequency and amount based on leaf color changes and soil surface moisture distribution characteristics to ensure sufficient but not excessive water supply during key crop growth stages (such as the heading stage). Regarding soil structure improvement, it is recommended to adopt measures such as deep tillage and loosening of the soil, and the addition of soil conditioners to reduce the density of the soil profile and improve soil aeration and permeability, thereby promoting root growth and expansion. Through these steps, a comprehensive, in-depth phenotypic identification report is ultimately generated, containing information on the correlation between phenotypic characteristics and the saline-alkali land environment.

[0195] For example, the method further includes: step S160: extracting the crop visual image, the saline-alkali land environment visual image and the spatial position correspondence between the two corresponding to each associated time node from the set of time-related visual image pairs; extracting the crop environment interaction feature content and feature change distance of each feature node from the evolution trajectory of crop environment interaction features; and extracting crop growth status features, crop morphological structure features and physiological response features of crops to saline-alkali land environment from the crop phenotypic recognition results.

[0196] After generating the crop phenotypic recognition report, in order to further optimize and validate the phenotypic recognition model, it is necessary to construct a sample set using existing data for model training. First, extract the crop visual image, the saline-alkali land environment visual image, and the spatial correspondence between them corresponding to each associated time node from the set of temporally associated visual image pairs. The above data is the original image data and spatial correlation information, which forms the basis for constructing the sample.

[0197] Simultaneously, crop environment interaction feature content for each feature node is extracted from the evolutionary trajectory of crop environment interaction features. This content is a high-level feature representation after feature extraction and interaction mapping. The feature change distance between each feature node is also extracted, reflecting the degree of feature change over time. Furthermore, crop growth status features, morphological structure features, and physiological response features to saline-alkali soil environments are extracted from the crop phenotypic recognition results. These features will serve as label data for the samples.

[0198] Step S161: Integrate the extracted crop visual images, saline-alkali land environment visual images, spatial location correspondence, crop-environment interaction features, feature change distance, and crop phenotypic recognition results to form an original sample set, wherein each sample unit contains all of the above information under a single associated time node.

[0199] The extracted data are then integrated according to their associated time points. For each associated time point, the crop visual image, saline-alkali land environment visual image, spatial location correspondence, crop-environment interaction features, feature change distance (distance from the previous node), and crop phenotypic recognition results are combined to form an independent sample unit. All sample units from associated time points are combined to form the original sample set. Each sample unit contains complete information for a single associated time point.

[0200] Step S162: Label each sample unit in the original sample set. The labeling content includes the crop growth status feature labeling data, crop morphological structure feature labeling data, and crop physiological response feature labeling data corresponding to the sample unit. The labeling is based on the specific content of each feature in the crop phenotypic recognition results.

[0201] Based on the specific content of each feature in the crop phenotypic recognition results, each sample unit in the original sample set is labeled. For crop growth status feature labeling data, the labeling is based on the level or quantitative value of the leaf, stem, and root growth status as indicated in the recognition results; for crop morphological structure feature labeling data, the labeling is based on the specific parameter values ​​of the morphological structure features (such as leaf length-to-width ratio, stem diameter, number of root branches, etc.); for crop physiological response feature labeling data to the saline-alkali environment, the labeling is based on the type and degree of the physiological response features (such as response sensitivity level, adaptability level, etc.). Through labeling, each sample unit has clear supervisory information.

[0202] Step S163: Based on the chronological order of the associated time nodes, divide the labeled original sample set into a training sample set and a validation sample set. During the division process, maintain the temporal continuity of the sample units within each set. After the division, the training sample set and the validation sample set contain sample units corresponding to associated time nodes in different time periods.

[0203] Based on the chronological order of the associated time nodes, the labeled original sample set is divided into a training sample set and a validation sample set. To ensure that the time series characteristics of the samples are not destroyed during training and validation, the time series continuity of sample units within each set is maintained during the partitioning. For example, the sample units corresponding to the first 70% of associated time nodes are assigned to the training sample set, and these sample units correspond to time periods from the earliest time node to an intermediate time node; the sample units corresponding to the last 30% of associated time nodes are assigned to the validation sample set, and the corresponding time periods are from an intermediate time node to the latest time node. This partitioning ensures that the training and validation sample sets contain data from different time periods, effectively evaluating the model's ability to identify crop phenotypes at different growth stages.

[0204] Step S164: Perform feature integration and normalization on each sample unit in the training sample set. Associate and integrate the color features, texture features, and morphological features in the crop environment interaction feature content with the environmental visual features in the corresponding saline-alkali land environment visual image. Normalize all integrated features to generate the input feature vector for each sample unit. The dimension of the input feature vector is consistent with the number of parameters of the crop environment interaction feature content and the environmental visual features.

[0205] Feature processing is performed on each sample unit in the training sample set. First, the color, texture, and morphological features of the crop-environment interaction features in the sample unit are associated and integrated with the environmental visual features in the corresponding saline-alkali land environment visual image. For example, leaf color features are associated with soil surface moisture distribution features, and stem texture features are associated with soil profile structure features, etc., to form a comprehensive feature set.

[0206] Then, all integrated features are normalized. The min-max normalization method is used to map the value of each feature to the [0,1] interval, eliminating differences in units and numerical ranges between different features. After normalization, all features are arranged in a specific order to generate the input feature vector for each sample unit. The dimension of the input feature vector is equal to the total number of parameters of the integrated features, i.e., the sum of the number of parameters of the crop-environment interaction feature content and the number of parameters of the environmental visual features.

[0207] Step S165: Initialize the parameters of the AI ​​model. Set the input layer dimension of the AI ​​model to the dimension of the input feature vector, and set the output layer dimension to the number of feature annotation data in the label annotation. The number of hidden layers and the number of neurons in each layer are determined according to the dimension of the input feature vector and the dimension of the output layer.

[0208] Choose a suitable AI model architecture, such as a hybrid model combining a multilayer perceptron (MLP) or convolutional neural network (CNN) with a recurrent neural network (RNN). Considering that the data includes both temporal and image features, a hybrid model may be more appropriate. Initialize the model parameters: set the input layer dimension to the dimension of the input feature vector to ensure that the input features can be smoothly input into the model; set the output layer dimension to the number of feature annotations in the label annotation, i.e., the total number of labeled parameters for the three types of features: crop growth status, morphological structure, and physiological response, so that the model can output complete phenotypic feature prediction results.

[0209] The number of hidden layers and the number of neurons per layer are determined based on the dimensions of the input feature vector and the output layer. Generally, when the dimension of the input feature vector is high, the number of hidden layers and the number of neurons per layer can be appropriately increased to enhance the model's feature learning ability. For example, if the dimension of the input feature vector is D and the dimension of the output layer is O, three hidden layers can be set, with the first layer having 2D neurons, the second layer having D neurons, and the third layer having O / 2 neurons, thus extracting key features through progressive dimensionality reduction.

[0210] Step S166: Input the input feature vectors from the training sample set into the AI ​​model one by one. After each input, obtain the predicted crop phenotypic features output by the AI ​​model. Compare the predicted crop phenotypic features with the feature annotation data of the sample unit and calculate the degree of difference between the two.

[0211] The input feature vectors from the training sample set are sequentially input into the initialized AI model. The model performs forward propagation computation on each input feature vector, and after feature transformation and nonlinear mapping in each hidden layer, outputs the predicted crop phenotypic features from the output layer. The predicted crop phenotypic features are then compared parameter-by-parameter with the feature annotation data of the sample units to calculate the degree of difference. The mean squared error (MSE) is used as a measure of this difference; it is calculated as the average of the squares of the differences between the predicted and labeled values. The smaller the MSE value, the closer the prediction result is to the true annotation.

[0212] Step S167: Adjust the parameters of the AI ​​model according to the degree of difference. During the adjustment process, refer to the input feature vector and the corresponding labeled feature data in the validation sample set. After each adjustment, input the input feature vector of the validation sample set into the AI ​​model to obtain the validation prediction result. Further adjust the parameters of the AI ​​model according to the degree of difference between the validation prediction result and the labeled feature data.

[0213] Based on the calculated mean variability (MSE value), the parameters of the AI ​​model are adjusted using the backpropagation algorithm. Starting from the output layer, the gradient of each layer's parameters with respect to the mean variability is calculated, and then the parameters are updated according to the gradient descent direction to reduce prediction variance. During parameter adjustment, the model is periodically validated using a validation sample set. The input feature vectors of the validation sample set are input into the currently adjusted model to obtain the validation prediction results, and the MSE value between the validation prediction results and the labeled feature data is calculated.

[0214] If the validation MSE value gradually decreases as the parameters are adjusted, it indicates that the model's generalization ability is improving, and the parameters should continue to be adjusted in the current direction. If the validation MSE value no longer decreases or even increases, overfitting may occur. In this case, it is necessary to adjust the learning rate, increase the regularization term, or reduce the number of neurons in the hidden layer to further optimize the model parameters.

[0215] Step S168: Repeat the steps of inputting training samples, comparing differences, adjusting parameters, and verifying the adjustment effect until the degree of difference between the AI ​​model's prediction results on the training sample set and the labeled feature data meets the preset training termination condition, and the degree of difference between the AI ​​model's prediction results on the verification sample set and the labeled feature data meets the preset verification pass condition, thus obtaining the trained AI model.

[0216] The process involves repeatedly inputting training samples into the model, comparing prediction differences, adjusting model parameters, and verifying the adjustment effect. The preset training termination condition is that the predicted MSE value of the training sample set is less than a certain threshold, and this MSE value no longer decreases significantly after multiple consecutive training cycles. The preset validation pass condition is that the predicted MSE value of the validation sample set is less than another set threshold. When the model simultaneously meets both the training termination condition and the validation pass condition, training stops, resulting in a fully trained AI model. This model can accurately predict crop phenotypic characteristics based on the input crop environment interaction features and environmental visual features.

[0217] Step S169: Select the associated time nodes that did not participate in the division of the training sample set and the validation sample set from the associated time nodes, form an independent sample set according to the construction method of the original sample set, input the input feature vector of the independent sample set into the trained AI model, obtain the independent prediction results, compare the independent prediction results with the feature annotation data labeled by the independent sample set, calculate the prediction accuracy of the crop growth state feature annotation data, the prediction matching degree of the crop morphological structure feature annotation data, and the prediction consistency of the crop physiological response feature annotation data, and complete the performance evaluation of the trained AI model.

[0218] To comprehensively evaluate the performance of the trained AI model, relevant time nodes that were not involved in the division of the training and validation sample sets (i.e., time periods independent of the training and validation sets) were selected from the relevant time nodes. Following the construction method of the original sample set, the various types of data corresponding to the above time nodes were integrated to form an independent sample set.

[0219] The input feature vectors of independent sample sets are fed into the trained AI model to obtain independent prediction results. The independent prediction results are compared with the feature-annotated data of the independent sample sets to calculate the model's prediction accuracy (the proportion of accurately predicted samples out of the total samples), prediction matching degree (the similarity between predicted feature parameter values ​​and labeled values) for crop growth state feature annotation data, and prediction consistency (the degree of consistency between predicted response types and labeled response types) for crop physiological response feature annotation data. These metrics comprehensively evaluate the model's generalization ability and prediction performance, completing the performance evaluation of the AI ​​model. If the evaluation results meet the practical application requirements, the model can be used for subsequent crop phenotypic recognition work; otherwise, the model structure or training parameters need to be readjusted, and the model needs to be trained and evaluated again.

[0220] Throughout the data processing, the acquisition of crop and environmental visual images is involved, which may include information such as the terrain and soil of the shooting area, but does not involve sensitive personal privacy data. If privacy-sensitive data is involved in practical applications, data anonymization techniques can be used, such as blurring or deleting sensitive identifying information in the images, and encrypting the location information of the acquisition device during transmission and storage, to ensure data security during acquisition, transmission, and storage and prevent privacy leaks.

[0221] In one exemplary embodiment, a computer vision-based crop phenotypic recognition system is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2As shown, the computer vision-based crop phenotyping system includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a computer vision-based crop phenotyping method. The display unit of this computer vision-based crop phenotyping system is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of this computer vision-based crop phenotyping system can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad set on the casing of the computer vision-based crop phenotyping system, or an external keyboard, touchpad, or mouse, etc.

[0222] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A crop phenotypic recognition method based on computer vision, characterized in that, The method includes: Obtain the temporal visual sequence of crops and the corresponding temporal visual sequence of the saline-alkali land environment within the planting area of ​​saline-alkali land; By extracting visual images from the same time point in the time-series crop visual sequence and the time-series saline-alkali land environment visual sequence, the spatial positional correspondence between crop visual images and saline-alkali land environment visual images at the same time point is established, resulting in a set of time-related visual image pairs. Perform feature interaction extraction processing on each temporally associated visual image pair in the set to generate crop-environment interaction features; Based on the temporal variation patterns of crop visual sequences, we perform evolutionary trajectory modeling on the crop-environment interaction features at each time point, analyze the differences in changes of crop-environment interaction features at adjacent time points, and construct the evolutionary trajectory of crop-environment interaction features. Crop phenotypic identification results are generated based on the evolutionary trajectory of crop-environment interaction characteristics, and a phenotypic identification report containing information on the association between phenotypic characteristics and saline-alkali land environment is output based on the crop phenotypic identification results. The step of performing feature interaction extraction processing on each time-related visual image pair in the set of time-related visual image pairs to generate crop-environment interaction features includes: Extract the first temporally associated visual image pair from the set of temporally associated visual image pairs. This temporally associated visual image pair contains crop visual images, saline-alkali land environment visual images, and the spatial correspondence between the two. Organ region segmentation is performed on crop visual images. Based on the color differences and morphological contours of crop organs, the crop visual images are divided into leaf regions, stem regions, and root regions. Visual features are extracted separately for each region. For the leaf area, extract the color distribution features, texture arrangement features and edge contour features of the leaf area. The color distribution features reflect the distribution ratio of different colors in the leaf area, the texture arrangement features reflect the arrangement pattern of the texture on the leaf surface, and the edge contour features reflect the shape features of the leaf contour. For the stem region, the color uniformity feature, texture density feature and morphological structure feature of the stem region are extracted. The color uniformity feature reflects the uniformity of the color in the stem region, the texture density feature reflects the density of the texture on the stem surface, and the morphological structure feature reflects the thickness and curvature of the stem. For the root system region, extract the color depth features, texture direction features, and branch distribution features of the root system region. The color depth features reflect the color depth changes of the root system region, the texture direction features reflect the extension direction of the root texture, and the branch distribution features reflect the distribution of root branches. Environmental region segmentation is performed on the visual image of saline-alkali land environment. Based on the texture differences and structural features of the environmental regions, the visual image of saline-alkali land environment is divided into soil surface region, soil profile region and irrigation region. Environmental visual features are extracted separately for each region. For the soil surface area, extract the color change characteristics, texture roughness characteristics, and crack distribution characteristics of the soil surface area; for the soil profile area, extract the layer distribution characteristics, color transition characteristics, and structural density characteristics of the soil profile area; for the irrigated area, extract the water distribution characteristics, color wetness characteristics, and boundary clarity characteristics of the irrigated area. The visual features of the leaf region, stem region, root region, soil surface region, soil profile region, and irrigation region were normalized respectively. Based on the spatial correspondence in the temporally associated visual image pairs, the normalized visual features of the leaf region were mapped and associated with the normalized environmental visual features of the soil surface region, the normalized visual features of the stem region were mapped and associated with the normalized environmental visual features of the soil profile region, and the normalized visual features of the root region were mapped and associated with the normalized environmental visual features of the irrigation region. All association results were integrated to generate the crop environment interaction features of the first temporally associated visual image pair. All time-related visual image pairs in the time-related visual image pair set are processed sequentially to obtain the crop-environment interaction features corresponding to each time-related visual image pair.

2. The crop phenotypic recognition method based on computer vision according to claim 1, characterized in that, The acquisition of temporal crop visual sequences and corresponding temporal saline-alkali land environment visual sequences within the saline-alkali land planting area includes: Determine the crop planting area and environmental monitoring area of ​​the saline-alkali land planting area. After adjusting the environmental monitoring area to fully cover the crop planting area, adjust the acquisition parameters of the relevant visual acquisition equipment to adapt the acquisition parameters to the light conditions of the saline-alkali land environment and the height characteristics of crop growth. The acquisition parameters include acquisition resolution, acquisition frame rate and exposure time. A visual acquisition time plan was constructed, and the acquisition interval was determined according to the crop growth cycle. At each acquisition time point, visual images of the crop within the planting area were acquired first. During acquisition, multiple angles were taken for the leaf parts, stem parts and root parts of the crop. Through multi-angle shooting, the visual images of each part can present the shape and color characteristics of that part. After completing the crop visual image acquisition at the same acquisition time point, immediately acquire the environmental visual image of the saline-alkali land within the corresponding environmental monitoring range. During the acquisition, fixed-point shooting is carried out on the soil surface, soil profile and irrigation area respectively. Through fixed-point shooting, the visual image of each environmental area can fully present the texture and structural features of the environmental area. The leaf, stem, and root visual images obtained at each collection time point are classified into crop visual image groups for that collection time point. All crop visual image groups at all time points are arranged in chronological order of collection time to form a time-series crop visual sequence. During the arrangement process, the specific collection time information corresponding to each crop visual image group is recorded. The visual images of the soil surface, soil profile, and irrigation area obtained at each collection time point are classified into a visual image group of the saline-alkali land environment at that collection time point. The visual image groups of the saline-alkali land environment at all time points are arranged in the order of collection time to form an initial temporal visual sequence of the saline-alkali land environment. By comparing the acquisition time information of the time-series crop visual sequence and the initial time-series saline-alkali land environment visual sequence, the arrangement order of each saline-alkali land environment visual image group in the initial time-series saline-alkali land environment visual sequence is adjusted so that the acquisition time of each saline-alkali land environment visual image group is completely consistent with the acquisition time of the corresponding crop visual image group in the time-series crop visual sequence. It was confirmed that each group of saline-alkali land environment visual images in the adjusted time-series saline-alkali land environment visual sequence includes soil surface visual images, soil profile visual images, and irrigation area visual images, and that the clarity and integrity of each visual image meet the requirements of subsequent processing. Finally, a time-series saline-alkali land environment visual sequence that corresponds to the time sequence of the time-series crop visual sequence was obtained.

3. The crop phenotypic recognition method based on computer vision according to claim 1, characterized in that, The process involves extracting visual images from the same time points in the temporal crop visual sequence and the temporal saline-alkali land environment visual sequence, establishing the spatial correspondence between crop visual images and saline-alkali land environment visual images at the same time point, and obtaining a set of time-related visual image pairs, including: Extract the acquisition time nodes corresponding to all crop visual image groups from the time-series crop visual sequence and the time-series saline-alkali land environment visual sequence to form a crop time node list and an environment time node list. The crop time node list is matched with the environmental time node list to filter out time nodes with completely identical collection time information. The filtered time nodes are identified as associated time nodes, forming an associated time node list. For the first associated time node in the associated time node list, extract the crop visual image group corresponding to the associated time node from the time-series crop visual sequence. The crop visual image group includes leaf visual images, stem visual images and root visual images. Extract the visual image set of the saline-alkali land environment corresponding to the associated time node from the temporal visual sequence of the saline-alkali land environment. The visual image set of the saline-alkali land environment includes visual images of the soil surface, visual images of the soil profile, and visual images of the irrigation area. Spatial coordinate calibration was performed on the crop visual image group and the saline-alkali land environment visual image group under the associated time node. Using the fixed marker in the crop planting area as a reference, the spatial coordinates of the fixed marker in the crop visual image and the saline-alkali land environment visual image were determined respectively. The spatial coordinates of the fixed marker in the two sets of images were kept consistent through coordinate transformation. Based on the calibrated spatial coordinates, the spatial positional correspondence between the visual images of leaves and soil surfaces, the spatial positional correspondence between the visual images of stems and soil profiles, and the spatial positional correspondence between the visual images of roots and the visual images of irrigated areas are established at the associated time node, forming a time-related visual image pair at the associated time node. All associated time nodes in the associated time node list are processed sequentially to generate a time-related visual image pair for each associated time node. All generated time-related visual image pairs are integrated to obtain a set of time-related visual image pairs. Each element in the set of time-related visual image pairs contains associated time node information, crop visual image, saline-alkali land environment visual image, and the spatial location correspondence between the two.

4. The crop phenotypic recognition method based on computer vision according to claim 1, characterized in that, The method, based on the temporal variation patterns of crop visual sequences, performs evolutionary trajectory modeling on the crop-environment interaction features at each time point, analyzes the differences in changes of crop-environment interaction features between adjacent time points, and constructs the evolutionary trajectory of crop-environment interaction features, including: Extract the arrangement order of all time nodes from the visual sequence of time-series crops, determine the time interval between each time node, and form a time node sequence that reflects the temporal variation pattern of crop growth. Crop-environment interaction features corresponding to each time node are extracted from the set of time-related visual image pairs. The extracted crop-environment interaction features are arranged in the order of the time node sequence to form a crop-environment interaction feature sequence. The first crop-environment interaction feature in the crop-environment interaction feature sequence is selected as the initial feature point, and the time node information and feature composition of the corresponding initial feature point are recorded. The second crop environment interaction feature in the crop environment interaction feature sequence is selected as the subsequent feature point. The subsequent feature point is compared with the initial feature point to analyze the differences in color features, texture features and morphological features. For color features, calculate the change in the proportion of color distribution in the corresponding regions between the initial feature points and subsequent feature points to determine the direction and magnitude of color feature changes; for texture features, analyze the changes in the texture arrangement pattern or density in the corresponding regions between the initial feature points and subsequent feature points to determine the trend of texture feature changes; for morphological features, compare the changes in the contour shape or structure of the corresponding regions between the initial feature points and subsequent feature points to determine the morphological feature change pattern. Based on the differences in color, texture and morphological features, the feature change distance between the initial feature point and the subsequent feature point is calculated. This feature change distance reflects the degree of change in the crop environment interaction features at two time points. Starting from the initial feature point and taking subsequent feature points as the next node, draw a feature change line segment between two points based on the feature change distance and the corresponding time interval. The length of the feature change line segment corresponds to the feature change distance, and the extension direction of the feature change line segment corresponds to the feature change trend. Using the second crop-environment interaction feature as the new initial feature point, and selecting the third crop-environment interaction feature as the new subsequent feature point, repeat the steps of feature comparison, change difference analysis, feature change distance calculation, and feature change line segment drawing. Process all crop-environment interaction features in the crop-environment interaction feature sequence in sequence, and connect all feature change lines in sequence to form an evolutionary trajectory that reflects the changes of crop-environment interaction features over time. Each node in this evolutionary trajectory contains information about the corresponding time node, the content of the crop-environment interaction feature, and the change relationship with the previous node.

5. The crop phenotypic recognition method based on computer vision according to claim 1, characterized in that, The generation of crop phenotypic recognition results based on the evolutionary trajectory of crop-environment interaction features includes: The evolution trajectory of crop-environment interaction features is decomposed into nodes, and each feature node in the evolution trajectory is extracted. Each feature node contains time node information, crop-environment interaction feature content, and change relationship with the previous node. For the crop environment interaction feature content of each feature node, the crop organ visual feature part is separated, which includes the visual features of the leaf area, stem area and root area. By analyzing the visual features of the crop organs, we can obtain the characteristics of crop growth status, crop morphological structure, and the physiological response of crops to saline-alkali soil environment. By integrating crop growth status characteristics, crop morphological structure characteristics, and crop physiological response characteristics to saline-alkali soil environment, crop phenotypic identification results are obtained.

6. The crop phenotypic recognition method based on computer vision according to claim 5, characterized in that, The analysis of the visual characteristics of the crop organs yields characteristics of crop growth status, crop morphological structure, and physiological response of the crop to the saline-alkali soil environment, including: This study analyzes the changes in visual characteristics of leaf regions over time. By comparing the color distribution, texture arrangement, and edge contour features of leaf regions at different time points, it determines the growth rate and stability of the leaf's growth state. The study also statistically analyzes the proportion of green in the color distribution; when the green proportion meets a preset photosynthetic threshold, the leaf's photosynthetic capacity is considered to have reached the standard. Furthermore, it records the transformation of texture arrangement from disordered to regular, corresponding to the transition from leaf immaturity to maturity. Finally, it calculates the expansion value of the contour range in the edge contour features, and the ratio of this expansion value to the time interval corresponds to the quantification of the leaf's growth rate. All analytical information is then integrated to form the leaf growth state characteristics. This study analyzes the changes in visual characteristics of the stem region over time, comparing the color uniformity, texture density, and morphological features of the stem region at different time points. It calculates the numerical changes in color uniformity, determining that stem nutrient absorption has reached a balanced standard when the numerical changes meet a preset nutrient threshold. It also statistically analyzes the numerical increments of texture density, determining that stem lignification has reached a preset stage when the numerical increments meet a preset lignification threshold. Finally, it measures the numerical changes in stem diameter and the angle of stem deviation from the vertical direction, with diameter changes corresponding to growth thickness and angle values ​​corresponding to uprightness. All analytical information is then integrated to form a stem growth status characteristic. This study analyzes the changes in visual characteristics of the root system over time, comparing the color intensity, texture direction, and branch distribution characteristics of the root system at different time points. It also calculates the uniformity of color intensity, determining root vitality when the uniformity meets a preset vitality threshold. Furthermore, it records the direction of texture direction extending towards water-rich areas, corresponding to root hydrotropism. The study measures the increase in the number of branches and the expansion area of ​​the branch distribution, with the increase corresponding to branch growth and the expansion area corresponding to expansion capacity. Finally, it integrates all analytical information to form root growth status characteristics. Finally, it combines leaf, stem, and root growth status characteristics to obtain crop growth status characteristics. The morphologically relevant parts of the visual features of crop organs in each feature node are extracted. The edge contour features of the leaf region, the morphological structure features of the stem region, and the branch distribution features of the root region are all morphologically relevant parts. By comparing the edge contour features of the leaf region at different time points, data on the stage transition of leaf shape from slender to broad and round is recorded, corresponding to the growth stage results. The growth value of the contour area is calculated, corresponding to the size increase. The number of continuous, unbroken segments of the contour lines is counted; when the number of continuous, unbroken segments meets a preset integrity threshold, the contour integrity is deemed to have met the standard. Based on all the analysis information, leaf morphological and structural features are formed. By comparing the morphological and structural features of the stem region at different time points, the change in stem diameter is measured, corresponding to the thickness change results. The fluctuation range of the angle between the stem axis and the vertical line is calculated. When the range of motion meets the preset stability threshold, the overall morphological stability is determined to have reached the standard; stem morphological structure features are formed based on all analytical information; the branch distribution features of the root system region at different time points are compared, and the numerical changes in the number of branches are statistically analyzed, with the numerical changes corresponding to the branch number changes; the numerical changes in branch length are measured, with the numerical changes corresponding to the branch length changes; the area changes of the root system coverage area are calculated, with the area changes corresponding to the overall distribution range; root morphological structure features are formed based on all analytical information; leaf morphological structure features, stem morphological structure features, and root morphological structure features are integrated to obtain crop morphological structure features. Extract the correlation and change between crop organ visual features and environmental visual features in each feature node, and analyze the response of crop organ visual features to changes in environmental visual features; when the water distribution characteristics of the soil surface area change, record the time interval for leaf color to change from dark green to light green, and compare this time interval with a preset water response threshold. If the time interval is less than the preset water response threshold, it is determined that the leaf response to soil water changes has reached a preset sensitivity standard; when the structural density characteristics of the soil profile area change, count the number of root branches penetrating the dense soil area, and compare this number with a preset structural response threshold. If this number is greater than the preset structural response threshold, it is determined that the root system response to soil structure changes has reached a preset capacity standard; when the water distribution characteristics of the irrigation area change, measure the speed at which root color changes from light to dark, and compare this speed with a preset absorption response threshold. If this speed is greater than the preset absorption response threshold, it is determined that the root system absorption response to irrigation water has reached a preset efficiency standard. By integrating the visual characteristics of crop organs in response to different environmental visual characteristics, the length of the cycle during which crops maintain normal growth under fluctuating soil moisture is statistically analyzed. When this cycle length meets a preset adaptation threshold, the crop's adaptability to water is determined to meet the standard. The range of root expansion of crops under dense soil structure is statistically analyzed. When this range meets a preset tolerance threshold, the crop's tolerance to soil structure is determined to meet the standard. Based on all the analyzed information, physiological response characteristics of crops to saline-alkali soil environment are formed.

7. The crop phenotypic recognition method based on computer vision according to claim 1, characterized in that, The phenotypic identification report, which outputs information on the correlation between phenotypic features and saline-alkali land environment based on the crop phenotypic identification results, includes: The crop growth status characteristics, crop morphological structure characteristics, and crop physiological response characteristics to saline-alkali land environment in the crop phenotypic recognition results are organized into independent feature modules. Each feature module includes a feature name, specific feature content, and the time node range corresponding to the feature. Extract the environmental visual images and environmental visual features of saline-alkali land corresponding to each time node from the set of time-related visual image pairs. Classify and organize the environmental visual features according to soil surface area, soil profile area and irrigation area to form an environmental feature module. The environmental feature module includes environmental area type, environmental visual feature content and corresponding time node information. Establish an association index between phenotypic features and environmental features. Based on the correspondence of time nodes, associate each sub-feature in crop growth status features with the environmental visual features at the same time node, associate each sub-feature in crop morphological structure features with the environmental visual features at the same time node, and associate each sub-feature in crop physiological response features with the environmental visual features that trigger the corresponding crop physiological response. Each association entry includes the phenotypic feature name, environmental feature name, association time node, and a description of the association changes. Construct a structural framework for the phenotypic identification report, which includes the report title, report generation time, data source description, overview of phenotypic features, overview of environmental features, phenotypic-environment correlation analysis, and conclusions and recommendations. In the overview of phenotypic features, the contents of the crop growth status feature module, crop morphological structure feature module, and crop physiological response feature module are presented in sequence. Each module displays feature changes in chronological order and is accompanied by corresponding crop visual image thumbnails. The environmental features overview section presents the content of the environmental features module, showing the changes in environmental visual features in the order of soil surface area, soil profile area and irrigation area, and is accompanied by corresponding thumbnail images of saline-alkali land environment. In the section on the association between phenotype and environment, all association index entries are displayed in chronological order of the association time points. Each entry details how phenotypic characteristics change with environmental characteristics, and also labels the corresponding visual image source. In the conclusion and recommendations section, based on the results of the phenotypic-environment association analysis, the advantages and disadvantages of crop growth in the current saline-alkali environment are summarized, and suggestions are made on soil moisture adjustment and soil structure improvement. Finally, a phenotypic identification report containing phenotypic characteristics and information on the association between saline-alkali environment is generated.

8. A crop phenotypic recognition system based on computer vision, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the computer vision-based crop phenotypic recognition method according to any one of claims 1 to 7 by executing the machine-executable instructions.

9. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. The processor of the computer vision-based crop phenotyping system reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the computer vision-based crop phenotyping system to perform the computer vision-based crop phenotyping method as described in any one of claims 1 to 7.

Images