A wheat disease and pest identification system based on video stream AI analysis

Abstract

The application provides a wheat disease and pest identification system based on video stream AI analysis, and relates to the field of electric digital data processing, comprising a data acquisition module, an archive video construction module and a disease and pest analysis identification module, the data acquisition module shoots image data at different time nodes of wheat growth, the archive video construction module extracts single plant image from the collected image data, creates an archive containing coding, spatial information and growth archive video for each wheat, the disease and pest analysis identification module analyzes the archive video, extracts disease and pest characteristics such as disease spot, color, texture and morphological anomaly, analyzes the continuity, development trend and spatial consistency of the characteristics in time sequence, and realizes disease and pest type identification and occurrence stage determination through time sequence evolution analysis; the application can identify wheat diseases and pests earlier and more stably, and provides a reliable basis for agricultural management.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing, and specifically to a wheat disease and pest identification system based on video stream AI analysis. Background Technology

[0002] Wheat is one of the world's most important food crops, and pests and diseases severely affect its yield and quality. Traditional wheat pest and disease identification relies mainly on manual field inspections, observing the color, spots, and morphology of wheat leaves to determine if pests or diseases have occurred. This method is not only inefficient but also requires a high level of expertise from the inspectors, making it difficult to achieve large-scale, precise pest and disease monitoring.

[0003] With the development of computer vision and artificial intelligence technologies, crop pest and disease detection methods based on image recognition and machine learning have gradually become a research hotspot. Existing technologies mainly achieve the identification or prediction of pests and diseases by analyzing images of crop plants or using biological data to build predictive models.

[0004] For example, Chinese patent CN121366639A discloses a method for predicting bacterial wilt disease in crops based on an ensemble learning model. This method includes the following steps: S1, collecting publicly available 16S rRNA gene sequences related to bacterial wilt in solanaceous crops and preprocessing the raw sequencing data using the standardized EasyAmplicon workflow; S2, using principal component analysis to reduce the data dimensionality while retaining 95% of the variance; S3, performing hyperparameter searches based on 5-fold cross-validation and grid search for the LightGBM, CatBoost, and XGBoost models respectively, selecting the three optimal sets of hyperparameters for each model; S4, constructing a model based on the three optimal sets of hyperparameters for each model and making predictions. However, this system is not suitable for monitoring crop pests and diseases based on images. First, identification or prediction methods based on single-point-of-time data struggle to fully reflect the dynamic development of pests and diseases. Whether it's visual recognition based on a single image frame or prediction methods based on biological data at a specific moment, they can only acquire state information at that particular instant, failing to capture the evolutionary patterns of pests and diseases over time. The occurrence and development of pests and diseases is a dynamic process, with characteristics exhibiting specific trends over time, such as gradually expanding lesion area, deepening color, and increasing severity of damage. Ignoring this temporal evolution information reduces the reliability and stability of identification. In precision agriculture applications, continuous observation of the growth status of the same crop at different times is necessary to promptly detect the occurrence of pests and diseases and assess their development trends. However, existing methods typically only allow isolated analysis of data from a single moment, failing to establish growth records for individual crops and hindering continuous tracking and trend analysis. This deficiency makes it difficult for existing technologies to meet the needs of early warning and precision control.

[0005] Furthermore, existing technologies are insufficient for identifying pests and diseases in their initial stages. In the early stages, pests and diseases often lack distinct characteristics, and a single observation may not capture enough information. However, if crops can be continuously observed and the temporal changes in characteristics can be analyzed, even if initial characteristics are weak, early identification can be achieved through accumulated information and trends at multiple time points, gaining valuable time for timely control measures.

[0006] Furthermore, existing methods lack the ability to accurately determine the stage of pest and disease outbreaks. In practical applications, it is necessary not only to identify the type of pest or disease, but also to determine whether it is in the initial, developing, or stable stage in order to take appropriate control measures. Methods based on single observations are insufficient to obtain dynamic information such as the rate of development and expansion trends, and therefore cannot accurately determine the stage of pest and disease outbreaks.

[0007] Therefore, there is a need for a technical solution that can fully utilize time-based information, establish crop growth records, and achieve continuous tracking, identification, and stage-based judgment of pests and diseases, in order to overcome the shortcomings of existing technologies, improve the accuracy, stability, timeliness, and practicality of crop pest and disease identification, and meet the development needs of precision agriculture and smart agriculture. Summary of the Invention

[0008] The purpose of this invention is to address the shortcomings by proposing a wheat disease and pest identification system based on video stream AI analysis.

[0009] The present invention adopts the following technical solution: A wheat disease and pest identification system based on video stream AI analysis includes a data acquisition module, an archive video construction module, and a disease and pest analysis and identification module. The data acquisition module is used to capture image data of wheat at different time points during wheat growth. The archive video construction module is used to extract individual plant images from the collected image data, and stitch the extracted individual plant images together according to the time sequence to obtain the growth archive video stream of each wheat plant. The pest and disease analysis and identification module is used to identify and identify pests and diseases in the archive video. Based on the wheat morphological changes reflected in the continuous frame images in the archive video, the module extracts lesion-related features, color change features, texture change features, and pest-related features, and analyzes the changing trends of the above features in the time dimension.

[0010] Furthermore, when the archive video construction module extracts the image region corresponding to a single wheat plant from the image data, it first performs plant region determination processing on the image to distinguish the wheat plant region from the background region. On this basis, it distinguishes individual wheat plants within the plant region and locates the boundary of the target single plant to determine the spatial range of the single wheat plant. Subsequently, it extracts the corresponding single plant image region from the original image according to the spatial range and performs size adjustment or alignment processing on the single plant image region to generate a single wheat plant image for subsequent processing.

[0011] Furthermore, the archive video construction module also performs quality verification on the individual wheat plant images. When an individual plant image is detected to be occluded, blurred, or incomplete, the individual plant image is replaced.

[0012] Furthermore, when extracting individual wheat plant images for the first time, an archive is created for each wheat plant. The archive records the corresponding wheat's code, spatial information, and growth archive video. When extracting individual plant images in the second and subsequent rounds, the archive video construction module performs cross-temporal association with the corresponding archive based on the spatial information. Successful association indicates that the individual plant image and the archive object are the same wheat plant.

[0013] Furthermore, after completing the cross-temporal association of individual wheat plant images, the archive video construction module evaluates the change amount between the newly extracted individual plant images and the last frame image in the growth video to characterize the degree of morphological change of individual wheat plants during growth. When the change amount meets the preset splicing trigger condition, the individual wheat plant image corresponding to the current time node is determined as a valid frame and used for the temporal reconstruction of the growth archive video. When the change amount does not meet the splicing trigger condition, the individual wheat plant image at the current time node is not included in the growth archive video.

[0014] Furthermore, after determining the valid frames, the archive video construction module generates a virtual frame image based on the valid frame image and the last frame image of the current growth video, and splices the virtual frame image and the newly acquired valid frame image together to the end of the growth archive video, thereby obtaining the updated growth archive video.

[0015] Furthermore, the pest and disease analysis and identification module parses the input growth record video stream, analyzes the frame sequence in the video in chronological order, and extracts the overall outline of a single wheat plant, leaf distribution area, and local morphological information in each frame. After completing the frame sequence parsing, the pest and disease analysis and identification module extracts pest and disease related features from each frame image. These features include lesion areas and their outlines, color change features, texture abnormalities, and morphological abnormalities such as damage, curling, or missing leaves. During feature extraction, the pest and disease analysis and identification module uses the overall growth status of a single wheat plant as a reference to distinguish morphological differences caused by normal growth or posture changes.

[0016] Furthermore, the pest and disease analysis and identification module performs temporal evolution analysis on the extracted pest and disease features in the time series. This analysis assesses the continuity of features by judging whether lesions or pest features continue to appear in adjacent frames, judges whether they show a unidirectional development trend by analyzing the direction of change of feature area, color depth or morphological degree over time, and judges whether they remain relatively concentrated by comparing the spatial distribution of features in consecutive frames. After completing the temporal evolution consistency analysis of pest and disease characteristics, the pest and disease analysis and identification module matches the extracted features and their evolutionary patterns with the preset pest and disease characteristic patterns to determine whether a single wheat plant is affected by disease or pest and to identify the specific pest and disease type. At the same time, it determines the occurrence stage of pests and diseases based on the occurrence range, expansion speed and degree of change of pest and disease characteristics in the time dimension.

[0017] The beneficial effects achieved by this invention are: 1. This invention achieves continuous tracking and recording of the wheat growth process by constructing a growth archive video of a single wheat plant. By acquiring images at different time points and extracting individual plant images, a complete archive containing encoding, spatial information, and growth video is created for each wheat plant. This enables continuous tracking of the state changes of the same wheat plant throughout its entire growth cycle, overcoming the limitation of existing technologies that can only perform single, isolated observations.

[0018] 2. This invention employs an effective frame determination mechanism based on change assessment, effectively reducing data redundancy in growth archive videos. By assessing the change between newly extracted individual plant images and the last frame of the archive video, new images are only included in the archive video when the change meets preset conditions. This avoids storing a large amount of redundant data, significantly improving data storage and processing efficiency while ensuring the integrity of key growth information.

[0019] 3. This invention improves the continuity and smoothness of growth archive videos through virtual frame generation technology. Virtual frame images are generated between adjacent valid frames. Through morphological benchmark alignment, non-growth perturbation constraint processing, and intermediate transition morphology construction, the archive video can continuously display the wheat growth process. This eliminates non-growth morphological differences caused by external factors such as wind and leaf movement, maintaining the natural smoothness of the video and providing high-quality video data for subsequent pest and disease analysis.

[0020] 4. This invention fully utilizes the evolutionary patterns of pests and diseases over time, significantly improving the accuracy and stability of identification. The pest and disease analysis and identification module, through temporal evolution analysis of pest and disease features in consecutive frame images, including feature continuity assessment, development trend judgment, and spatial consistency analysis, can effectively distinguish between genuine pest and disease features and occasional noise or environmental interference, avoiding the misjudgment problems that are prone to occur in single-frame image-based recognition methods.

[0021] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall structural framework of the present invention; Figure 2 This is a schematic diagram of the video archive construction process of this invention; Figure 3 This is a schematic diagram of the virtual frame generation process of the present invention; Figure 4 This is a schematic diagram of the pest and disease analysis and identification process of this invention; Figure 5 This is a schematic diagram of the terminal interactive interface (UI) of the present invention. Detailed Implementation

[0023] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated beforehand. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention. Example 1:

[0024] This embodiment provides a wheat disease and pest identification system based on video stream AI analysis, combined with... Figure 1 It includes a data acquisition module, an archive video construction module, and a pest and disease analysis and identification module; The data acquisition module is used to capture image data of wheat at different time points during wheat growth. The archive video construction module is used to extract individual plant images from the collected image data, and stitch the extracted individual plant images together according to the time sequence to obtain the growth archive video stream of each wheat plant. The pest and disease analysis and identification module is used to identify and identify pests and diseases in the archive video. Based on the wheat morphological changes reflected in the continuous frame images in the archive video, the module extracts lesion-related features, color change features, texture change features and pest-related features, and analyzes the changing trends of the above features in the time dimension. Combination Figure 2 The archive video construction module extracts the image region corresponding to a single wheat plant from the image data. First, it performs plant region determination processing on the image to distinguish the wheat plant region from the background region. On this basis, it distinguishes individual wheat plants within the plant region and locates the boundary of the target single plant to determine the spatial range of the single wheat plant. Subsequently, it extracts the corresponding single plant image region from the original image according to the spatial range and performs size adjustment or alignment processing on the single plant image region to generate a single wheat image for subsequent processing. In addition, the archive video construction module also performs quality verification on the single wheat plant image. When it detects that the single plant image is occluded, blurry, or incomplete, it replaces the single plant image.

[0025] When extracting individual wheat plant images for the first time, a profile is created for each wheat plant, which records the corresponding wheat's code, spatial information, and growth profile video. When the archive video construction module extracts single-plant images in the second round and thereafter, it performs cross-temporal association with the corresponding archive based on spatial information. Successful association indicates that the single-plant image and the archive object are the same wheat plant. The spatial information includes the shooting angle, direction, and the position of the single plant image within the overall image; After completing the cross-temporal association of individual wheat plant images, the archive video construction module evaluates the change amount between the newly extracted individual plant images and the last frame image in the growth video to characterize the degree of morphological changes that occur in individual wheat plants during growth. When the change amount meets the preset splicing trigger condition, the individual wheat plant image corresponding to the current time node is determined as a valid frame and used for the temporal reconstruction of the growth archive video. When the change amount does not meet the splicing trigger condition, the individual wheat plant image at the current time node is not included in the growth archive video, thereby ensuring the integrity of key changes in wheat growth while reducing data redundancy in the growth archive video. After determining the valid frames, the archive video construction module generates a virtual frame image based on the valid frame image and the last frame image of the current growth video. The virtual frame image and the newly acquired valid frame image are then stitched together at the end of the growth archive video to obtain an updated growth archive video. Through this process, the growth archive video can continuously display the wheat growth process and maintain the natural and smooth playback of the video. The frames in the growth record video are divided into real-shot images and virtual-generated images. The former are called valid frames, and the latter are called virtual frames. Now, for the newly extracted individual plant images With the last frame of the growth video The process of assessing changes will be elaborated in detail below: The archival video construction module calculates the local morphological gradient of wheat based on image pixels. : ; in, express The pixel information in the i-th row and j-th column of the image. express The pixel information in the i-th row and j-th column of the image. Represents the local weight coefficient. Indicates the directional weighting coefficient; The local weights are calculated according to the following formula: ; in, Represents the neighborhood set of pixel (i, j); The directional weighting coefficient is calculated according to the following formula: ; in, This represents the vector indicating the growth direction of the wheat plant. This represents the direction vector of the current pixel change; The archive video construction module calculates the growth difference determination quantity according to the following formula. : ; in, Represents the set of pixels of a wheat plant; When the growth difference determination value is greater than the set threshold, the single plant image... It was determined to be a valid frame; Combination Figure 3 In the generation of virtual frames, the archive video construction module first takes the last valid frame in the current growth archive video and the newly acquired valid frame as the processing objects, and performs morphological reference alignment processing on the two frames. The morphological reference alignment takes the overall structure of a single wheat plant as a reference. By uniformly constraining the position of the main stem, the distribution area of ​​the leaves, and the outline range of the plant, the two valid frames are made comparable in spatial position and overall posture, thereby reducing the impact of small changes in shooting angle and the overall sway of the plant on the generation of subsequent virtual frames. After completing the morphological benchmark alignment, the archive video construction module extracts the morphological difference information between the two effective frames before and after, and analyzes the change characteristics of the difference. Among them, local morphological abrupt changes caused by external factors such as wind blowing and leaf swinging are judged as non-growth morphological disturbances. Such disturbances are constrained in the virtual frame generation process to make the morphological changes present a continuous and monotonous evolution trend in the time dimension, and avoid repeated swinging or abrupt changes that do not conform to the growth law. Based on the completion of the morphological difference constraint processing, the archive video construction module uses the growth state represented by the two effective frames before and after as the morphological boundary condition to construct several intermediate transitional morphological states between the two effective frames. The intermediate transitional morphological states are used to represent the intermediate forms that a single wheat plant may experience during continuous growth. The direction of its morphological change is consistent with the growth trend reflected by the effective frames before and after, and the change range is limited by the morphological difference range between the effective frames before and after. Subsequently, the archival video construction module maps each intermediate transitional state to generate corresponding virtual frame images, so that the virtual frames present a gradual evolution relationship in time sequence. The generated virtual frames avoid introducing new structural changes in their overall form. The change process can not only smoothly transition the morphological differences between adjacent effective frames caused by non-growth disturbances, but also reflect the gradual evolution of real growth differences in the continuous frame sequence, thereby ensuring the balance between visual continuity and growth authenticity in the growth archival video. Combination Figure 4 The disease and pest analysis and identification module is used to identify and identify wheat diseases and pests based on the video stream of a single wheat plant's growth record. Its ultimate goal is to accurately determine whether a single wheat plant has been affected by diseases and pests, as well as the type and stage of the diseases and pests. This module takes the wheat growth process reflected by the continuous frame images in the growth record video as the analysis object, and makes full use of the continuity of the video in the time dimension to comprehensively analyze the appearance, development and change process of disease and pest characteristics, thereby avoiding the risk of misjudgment caused by relying solely on a single frame image. During the process, the pest and disease analysis and identification module first parses the input growth record video stream. This growth record video consists of valid frames and virtual frames. The module parses the frame sequence in the video in chronological order, extracting the overall outline of a single wheat plant, leaf distribution area, and local morphological information from each frame. Since virtual frames are used to smooth non-growth morphological differences between adjacent valid frames caused by external factors such as wind, virtual frames and valid frames participate together in subsequent temporal analysis, helping to maintain the continuity and stability of pest and disease characteristics in the video sequence. After completing the frame sequence parsing, the pest and disease analysis and identification module extracts pest and disease related features from each frame image. These pest and disease related features include lesion areas and their outlines, color change features, texture abnormal features, and morphological abnormal features such as damage, curling, or missing leaves. During the feature extraction process, the pest and disease analysis and identification module uses the overall growth status of a single wheat plant as a reference to distinguish and process morphological differences caused by normal growth or posture changes, thereby reducing the impact of non-growth disturbances on the results of pest and disease feature extraction. Subsequently, the pest and disease analysis and identification module performs temporal evolution analysis on the extracted pest and disease features. This analysis does not focus on whether there are isolated abnormal features in a certain frame, but rather focuses on the changing patterns of related features in consecutive frames. Specifically, the pest and disease analysis and identification module assesses the continuity of features by judging whether lesions or pest features continue to appear in multiple adjacent frames. It judges whether features show a unidirectional development trend by analyzing the direction of change of feature area, color depth, or morphological degree over time. At the same time, it judges whether features remain relatively concentrated by comparing the spatial distribution of features in consecutive frames, thereby eliminating artifacts caused by random noise or short-term disturbances. With the smoothing effect of virtual frames in the time dimension, the above temporal analysis can more clearly reflect the true evolution process of pest and disease features. After completing the temporal evolution consistency analysis of pest and disease characteristics, the pest and disease analysis and identification module identifies and classifies wheat pests and diseases based on the analysis results. By matching the extracted features and their evolutionary patterns with preset pest and disease characteristic patterns, the module can determine whether a single wheat plant is affected by disease or pests, and further identify the specific types of pests and diseases. At the same time, the module determines the occurrence stage of pests and diseases based on the occurrence range, expansion speed, and degree of change of pest and disease characteristics over time, including the initial stage, development stage, or stable stage. Finally, the pest and disease analysis and identification module outputs the pest and disease identification results of a single wheat plant and the corresponding occurrence stage information, thereby realizing continuous identification and identification of wheat pests and diseases based on growth archive videos. Through the above processing, the pest and disease analysis and identification module makes full use of the continuous characteristics of growth archive videos in the time dimension. While suppressing non-growth disturbances, it highlights the evolutionary law of real pest and disease characteristics. Compared with the recognition method based on single-frame images, it can identify wheat pests and diseases earlier and more stably, providing a reliable basis for subsequent agricultural management and control decisions. In the t-th frame image, the pest and disease feature state vector is defined as : ; in, This represents the normalized area characteristic of the lesion region. This indicates the intensity of color abnormality relative to a healthy baseline. Indicates the degree of texture anomaly. Indicates the degree of abnormality in the local leaf morphology; The pest and disease state occupancy function for frame t is defined by the following formula. : ; in, There are thresholds for the characteristics of pests and diseases; Given a time window length of N, the stability of pest and disease status is defined according to the following formula. : ; in, It is a growth continuity decay factor; The pest and disease evolution effectiveness function is defined according to the following formula. : ; in, This represents the vector representing the change in pest and disease characteristics in the k-th frame. This indicates a component-wise non-subtractive relation; Given a time window length of N, the consistency of pest and disease evolution is defined according to the following formula. : ; Let the set of wheat leaf numbers containing the pest and disease characteristics in frame t be . The spatial structure consistency is calculated according to the following formula. : ; in, A value of 1 indicates that two sets of numbers are completely identical, and a value of 0 indicates otherwise. The pest and disease analysis and identification module is based on , and as well as Perform a matching analysis. Example 2:

[0026] This embodiment should be understood to include at least all the features of any of the foregoing embodiments, and to further improve upon them; The hardware structure of the data acquisition module includes an image acquisition device, a storage device, and a communication device; The image acquisition device is used to capture images of wheat plants at different stages of wheat growth. The device is equipped with a high-resolution image sensor, which can capture detailed features of wheat plants, including leaf texture, color changes, and lesion morphology. The image acquisition device is also equipped with an optical lens assembly for adjusting the shooting distance and focal length to adapt to changes in wheat plant height and canopy structure at different growth stages. To ensure the stability of image quality, the image acquisition device is equipped with an ambient light compensation unit, which can automatically adjust exposure parameters according to natural lighting conditions to avoid image quality fluctuations caused by changes in lighting. The storage device is used to locally cache the acquired image data. This device uses a high-capacity storage medium that can hold long-term sequence image data. The communication device is responsible for transmitting the acquired image data to the archive video construction module. This device supports wired or wireless data transmission methods to ensure that the image data can be delivered to the subsequent processing stages in a timely and complete manner. The hardware structure of the archival video construction module includes an image processing unit, a feature extraction unit, a storage management unit, and a video generation unit; The image processing unit receives raw image data from the data acquisition module and performs preprocessing operations on the image, including noise filtering, contrast enhancement, and color correction. This unit is equipped with a dedicated image processing chip, which can efficiently complete pixel-level calculations and provide high-quality input data for subsequent single-plant extraction and feature analysis. The feature extraction unit is responsible for identifying and extracting image regions of individual wheat plants from the preprocessed image. This unit first separates the wheat plant regions in the image from the background regions through a plant region detector, then distinguishes multiple wheat individuals within the plant region through an individual recognizer, and determines the spatial range of the target individual plant through a boundary locator. The feature extraction unit is also equipped with a quality checker to detect whether there are problems such as occlusion, blurring or incompleteness in the extracted single-plant images. When a quality problem is detected, the quality checker triggers the image replacement process to ensure that the images entering the archive meet the quality standards. The storage management unit is responsible for creating and maintaining growth profiles for each wheat plant. This unit is equipped with a profile database to store the wheat's coding information, spatial location information, and growth profile video data. The storage management unit is also equipped with a time-series association processor, which matches newly extracted single-plant images with existing profiles based on spatial information to ensure that images of the same wheat plant at different time points can be correctly archived. The video generation unit is responsible for stitching together historical images of individual wheat plants in chronological order to generate a growth archive video. This unit is equipped with a change evaluator, which calculates the morphological differences between newly extracted individual plant images and the last frame of the archive video. When the differences meet preset conditions, the change evaluator marks the new image as a valid frame. The video generation unit is also equipped with a virtual frame generator, which generates virtual frame images of intermediate transition states based on the morphological difference information between adjacent valid frames, using a morphological interpolation algorithm. The virtual frames are then stitched together with the valid frames into the archive video, making the video present a continuous and smooth growth process. The hardware structure of the pest and disease analysis and identification module includes a video parsing unit, a feature recognition unit, a time-series analysis unit, and an identification decision unit; The video parsing unit receives the growth archive video stream from the archive video construction module and parses each frame in the video in chronological order to extract the overall outline of wheat, leaf distribution, and local morphological information from the frame images. This unit is equipped with a frame sequence buffer to temporarily store image data of multiple consecutive frames, providing data support for subsequent time-series analysis.

[0027] The feature recognition unit is responsible for extracting pest and disease-related features from the parsed frame images. This unit is equipped with a lesion detector, a color analyzer, a texture analyzer, and a morphological anomaly detector. The lesion detector is used to identify lesion areas on the leaf surface and extract the contour morphological features of the lesions. The color analyzer is used to detect the degree of deviation of the leaf color from the healthy baseline and identify color anomalies such as fading, yellowing, or browning caused by diseases. The texture analyzer is used to extract the texture features of the leaf surface and identify changes in texture roughness or texture pattern anomalies caused by diseases or pests. The morphological anomaly detector is used to identify morphological changes caused by pests or diseases, such as leaf damage, curling, and missing parts. When extracting pest and disease features, the feature recognition unit is equipped with a reference state comparator. This comparator uses the normal growth state of wheat as a reference baseline to distinguish morphological differences caused by normal growth or posture changes, avoiding misjudging normal growth changes as pest and disease features. The temporal analysis unit is responsible for analyzing the evolutionary patterns of extracted pest and disease features over time. This unit is equipped with a continuity evaluator to determine whether pest and disease features appear continuously across multiple frames, thus distinguishing between genuine pest and disease features and occasional noise. The unit also includes a trend analyzer to analyze the changing trends of the area, color depth, or morphological degree of pest and disease features over time, determining whether the features exhibit a unidirectional evolutionary pattern. Furthermore, the unit is equipped with a spatial consistency analyzer to compare the spatial distribution of pest and disease features across consecutive frames, determining whether the features remain relatively concentrated, thereby eliminating artifacts caused by random perturbations. By integrating the analysis results of continuity, trend, and spatial consistency, the unit comprehensively evaluates the temporal evolution consistency of pest and disease features. The identification and decision-making unit is responsible for the final identification and assessment of wheat diseases and pests based on the time-series analysis results. This unit is equipped with a feature matcher, which is used to match the extracted disease and pest features and their evolutionary patterns with a preset disease and pest feature pattern library, thereby determining whether a single wheat plant is affected by disease or pests and identifying the specific disease or pest type. The identification and decision-making unit is also equipped with a stage determiner, which determines the occurrence stage of the disease or pest based on the occurrence range, expansion speed, and degree of change of the disease and pest features in the time dimension, including the initial stage, development stage, or stable stage. The identification and decision-making unit finally outputs the disease and pest identification results of a single wheat plant and the corresponding occurrence stage information, providing a basis for agricultural management and control decisions. Combination Figure 5 It serves as the interactive UI interface for the monitoring terminal, allowing users to select specific plants to view archive videos and exchange information on pest and disease status. In practical implementation, the image acquisition device of the data acquisition module can be an industrial-grade digital camera or an intelligent device equipped with an image sensor. The resolution of the image sensor should be sufficient to clearly capture the detailed features of wheat leaves. The optical lens assembly can be an adjustable focus lens or a zoom lens to adapt to different shooting distances and field of view requirements. The ambient light compensation unit can be implemented through an automatic exposure control circuit, which automatically adjusts the exposure time and gain parameters according to the light intensity received by the image sensor. The storage device can be a solid-state drive or a high-capacity memory card, with a storage capacity sufficient to hold image data for at least one complete growth cycle. The communication device can be an Ethernet interface, a wireless network module, or a mobile communication module, with the appropriate communication method selected based on the actual deployment environment. The image processing unit of the archival video construction module can use a digital signal processor or a graphics processor to provide sufficient image processing capabilities. The plant region detector in the feature extraction unit can be implemented based on a color space segmentation algorithm or a deep learning semantic segmentation model. The individual identifier can be implemented using an instance segmentation algorithm or a method based on connected component analysis. The boundary locator can determine the boundary of a single wheat plant through a contour detection algorithm. The quality checker can be implemented based on image sharpness evaluation indicators, occlusion detection algorithms, and integrity detection algorithms. The archival database of the storage management unit can use a relational database or a document database to structure and store wheat archival information. The time-series association processor can be implemented based on a spatial location matching algorithm or a feature similarity matching algorithm. The change evaluator in the video generation unit can be implemented based on image difference measurement indicators, including pixel-level differences, structural similarity, or feature change indicators. The virtual frame generator can be implemented using a morphological interpolation algorithm, an optical flow estimation algorithm, or an image generation model based on a generative adversarial network. The video parsing unit of the pest and disease analysis and identification module can be implemented using a video decoder or frame extractor. The frame sequence buffer can be implemented using a circular buffer or queue data structure. The lesion detector in the feature recognition unit can be implemented based on color space threshold segmentation, region growing algorithm or deep learning target detection model. The color analyzer can be implemented based on color space conversion and color statistical feature extraction. The texture analyzer can be implemented based on gray-level co-occurrence matrix, local binary mode or texture features extracted by convolutional neural network. The morphological anomaly detector can be implemented based on contour shape analysis, morphological operation or deep learning instance segmentation model. The reference state comparator can be implemented by establishing a morphological feature library of healthy wheat and calculating the deviation between the feature to be detected and the reference feature. The continuity evaluator in the time series analysis unit can be implemented based on the frequency statistics of features within the time window; the trend analyzer can be implemented based on time series analysis methods, including moving average, trend fitting, or rate of change calculation; and the spatial consistency analyzer can be implemented based on time series clustering analysis or spatial overlap calculation of feature location coordinates. The feature matcher in the identification decision unit can be based on pattern recognition algorithms, including nearest neighbor classifiers, support vector machines, or deep learning classification models, while the stage determiner can be based on feature evolution speed, cumulative change, and expert knowledge rules.

[0028] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A wheat disease and pest identification system based on video stream AI analysis, characterized in that, It includes a data acquisition module, an archive video construction module, and a pest and disease analysis and identification module; The data acquisition module is used to capture image data of wheat at different time points during wheat growth. The archive video construction module is used to extract individual plant images from the collected image data, and stitch the extracted individual plant images together according to the time sequence to obtain the growth archive video stream of each wheat plant. The pest and disease analysis and identification module is used to identify and identify pests and diseases in the archive video. Based on the wheat morphological changes reflected in the continuous frame images in the archive video, the module extracts lesion-related features, color change features, texture change features, and pest-related features, and analyzes the changing trends of the above features in the time dimension.

2. The wheat disease and pest identification system based on video stream AI analysis as described in claim 1, characterized in that, When the archive video construction module extracts the image region corresponding to a single wheat plant from the image data, it first performs plant region determination processing on the image to distinguish the wheat plant region from the background region. On this basis, it distinguishes individual wheat plants within the plant region and locates the boundary of the target single plant to determine the spatial range of the single wheat plant. Subsequently, it extracts the corresponding single plant image region from the original image according to the spatial range and performs size adjustment or alignment processing on the single plant image region to generate a single wheat image for subsequent processing.

3. The wheat disease and pest identification system based on video stream AI analysis as described in claim 2, characterized in that, The archive video construction module also performs quality verification on the individual wheat plant images. When an individual plant image is detected to be occluded, blurry, or incomplete, the individual plant image is replaced.

4. The wheat disease and pest identification system based on video stream AI analysis as described in claim 3, characterized in that, When extracting individual wheat plant images for the first time, an archive is created for each wheat plant. The archive records the corresponding wheat's code, spatial information, and growth archive video. When extracting individual plant images in the second and subsequent rounds, the archive video construction module performs cross-temporal association with the corresponding archive based on the spatial information. Successful association indicates that the individual plant image and the archive object are the same wheat plant.

5. The wheat disease and pest identification system based on video stream AI analysis as described in claim 4, characterized in that, After completing the cross-temporal association of individual wheat plant images, the archive video construction module evaluates the change amount between the newly extracted individual plant images and the last frame image in the growth video to characterize the degree of morphological changes that occur in individual wheat plants during growth. When the change amount meets the preset splicing trigger condition, the individual wheat plant image corresponding to the current time node is determined as a valid frame and used for the temporal reconstruction of the growth archive video. When the change amount does not meet the splicing trigger condition, the individual wheat plant image at the current time node is not included in the growth archive video.

6. The wheat disease and pest identification system based on video stream AI analysis as described in claim 5, characterized in that, After determining the valid frames, the archive video construction module generates a virtual frame image based on the valid frame image and the last frame image of the current growth video, and splices the virtual frame image and the newly acquired valid frame image together to the end of the growth archive video to obtain the updated growth archive video.

7. The wheat disease and pest identification system based on video stream AI analysis as described in claim 6, characterized in that, The pest and disease analysis and identification module parses and processes the input growth record video stream, analyzes the frame sequence in the video in chronological order, and extracts the overall outline of a single wheat plant, leaf distribution area and local morphological information in each frame. After completing the frame sequence parsing, the pest and disease analysis and identification module extracts pest and disease related features from each frame image. These features include lesion areas and their outlines, color change features, texture abnormalities, and morphological abnormalities such as damage, curling, or missing leaves. During feature extraction, the pest and disease analysis and identification module uses the overall growth status of a single wheat plant as a reference to distinguish morphological differences caused by normal growth or posture changes.

8. The wheat disease and pest identification system based on video stream AI analysis as described in claim 7, characterized in that, The pest and disease analysis and identification module performs temporal evolution analysis on the extracted pest and disease features. This analysis assesses the continuity of features by judging whether lesions or pest features continue to appear in adjacent frames, and judges whether they show a unidirectional development trend by analyzing the direction of change of feature area, color depth or morphological degree over time. At the same time, it judges whether they remain relatively concentrated by comparing the spatial distribution of features in consecutive frames. After completing the temporal evolution consistency analysis of pest and disease characteristics, the pest and disease analysis and identification module matches the extracted features and their evolutionary patterns with the preset pest and disease characteristic patterns to determine whether a single wheat plant is affected by disease or pest and to identify the specific pest and disease type. At the same time, it determines the occurrence stage of pests and diseases based on the occurrence range, expansion speed and degree of change of pest and disease characteristics in the time dimension.

Images