Seed disease granule recognition method based on image recognition and intelligent screening system

Abstract

The present application relates to the technical field of image recognition, solves the problem that it is difficult to accurately determine the seeds with unobvious disease characteristics or between normal and disease in the prior art while maintaining the seed disease kernel recognition efficiency, and provides a seed disease kernel recognition method and intelligent screening system based on image recognition. The method comprises the following steps: obtaining original images of each seed to be detected; preprocessing each original image to obtain each first target image; performing preliminary identification on each first target image to obtain an initial seed recognition result corresponding to each seed to be detected; when the initial seed recognition result is a seed to be reviewed, extracting the first target image corresponding to the seed to be reviewed as a second target image; performing secondary identification on the second target image to obtain a target seed recognition result corresponding to the seed to be reviewed; and determining the disease kernel recognition result of each seed to be detected. The present application realizes the consideration of seed disease kernel recognition efficiency and recognition accuracy.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method for identifying diseased seeds and an intelligent screening system based on image recognition. Background Technology

[0002] Seeds are the foundation of agricultural production and crop cultivation, and their quality directly affects crop germination rate, growth uniformity, and final yield. Diseased seeds typically exhibit problems such as mold, rot, insect damage, or tissue abnormalities. If mixed with normal seeds, they not only reduce sowing quality but may also trigger disease transmission, causing large-scale yield reductions. Therefore, accurately identifying and removing diseased seeds before processing, storage, and sowing is a crucial step in ensuring seed quality and agricultural production safety.

[0003] With the development of machine vision and image processing technologies, image recognition-based seed detection methods are increasingly being applied in the field of seed quality inspection. Existing technologies typically acquire seed images and utilize color analysis, grayscale analysis, and morphological feature extraction to inspect the seed appearance and determine the presence of disease characteristics. Some solutions also incorporate threshold segmentation, connected component analysis, or simple classification models to achieve automatic identification of diseased seeds, thereby replacing manual visual inspection and improving detection efficiency.

[0004] However, in practical applications, seed lesion characteristics often exhibit diversity and uncertainty. Different seed species show significant differences in appearance, and the severity of lesions varies. Existing image processing methods based on a single recognition strategy typically use uniform recognition rules or thresholds for judgment, which can easily lead to misclassification in seeds with indistinct lesion characteristics or those falling between normal and lesion. On the one hand, they may easily misclassify slightly lesioned seeds as normal seeds, affecting screening accuracy; on the other hand, they may also misclassify normal seeds with natural differences in appearance as lesioned seeds, causing unnecessary losses. Therefore, existing technologies still have shortcomings in terms of recognition accuracy and stability.

[0005] Therefore, how to improve the accuracy of identifying seeds with inconspicuous lesion characteristics while ensuring identification efficiency, and how to reliably determine the lesion of seeds, is a core technical problem that urgently needs to be solved in the existing technology. Summary of the Invention

[0006] In view of this, the present invention provides a seed lesion identification method and intelligent screening system based on image recognition, which solves the problem in the prior art that it is difficult to accurately determine seeds with indistinct lesion characteristics or those between normal and lesion while maintaining the efficiency of seed lesion identification.

[0007] The technical solution adopted in this invention is:

[0008] In a first aspect, the present invention provides a method for identifying seed lesions based on image recognition, the method comprising:

[0009] Acquire the original images of each seed to be tested under preset lighting conditions;

[0010] The original images are preprocessed to obtain the preprocessed first target images.

[0011] According to the preset first seed image recognition strategy, each of the first target images is initially identified to obtain the initial seed recognition result corresponding to each seed to be detected, wherein the initial seed recognition result includes diseased granules, granules to be re-examined, and normal granules;

[0012] When the initial seed identification result is a nucleate to be verified, the first target image corresponding to the nucleate to be verified is extracted as the second target image;

[0013] According to the preset second seed image recognition strategy, the second target image is re-identified to obtain the target seed recognition result corresponding to the particle to be verified, wherein the target seed recognition result includes diseased particles and normal particles;

[0014] Based on the initial seed identification results and the target seed identification results, the diseased grain identification results of each of the seeds to be tested are determined.

[0015] Preferably, the preprocessing of each of the original images to obtain the preprocessed target image includes:

[0016] Brightness compensation processing is performed on each of the original images, and contrast adjustment is performed on the compensated images to obtain each corrected image after brightness and contrast adjustment;

[0017] Based on the pixel difference between the seed contour and the background region in each corrected image, seed region extraction processing is performed on each corrected image to obtain various sub-region images;

[0018] Each of the seed region images is cropped and converted in image format to obtain the first target image.

[0019] Preferably, the step of performing preliminary identification on each of the first target images according to a preset first seed image recognition strategy to obtain the initial seed recognition result corresponding to each seed to be detected includes:

[0020] Foreground and background separation processing is performed on each of the first target images to extract the main body region corresponding to the seed to be detected, and the seed main body image is obtained;

[0021] Perform grayscale analysis or single color channel analysis on the seed body image to obtain the brightness distribution data or color distribution data of the pixels in the seed body image;

[0022] Based on a preset brightness threshold or color deviation threshold, and combined with the brightness distribution data or color distribution data, the seed main image is subjected to threshold segmentation processing to extract brightness abnormal regions or color abnormal regions, thereby obtaining a candidate set of abnormal regions.

[0023] The ratio of the pixel area of ​​the abnormal region in the candidate abnormal region set to the total pixel area of ​​the seed main image is calculated to obtain the abnormal region area ratio.

[0024] The area ratio of the abnormal region is compared with a preset area threshold range. Based on the comparison result, the seed to be detected is initially classified and determined to obtain the initial seed identification result.

[0025] Preferably, the step of performing foreground and background separation processing on each of the first target images to extract the main body region corresponding to the seed to be detected, and obtaining the seed main body image, includes:

[0026] Based on the brightness distribution features or color distribution features of pixels in each of the first target images, preliminary threshold segmentation processing is performed on each of the first target images to generate foreground candidate images for characterizing the seed region.

[0027] Based on the foreground candidate image, morphological processing is performed on the foreground region to obtain a processed foreground image that removes interference from non-seed regions. The morphological processing includes opening operations, closing operations, or noise region removal processing.

[0028] Perform connected component analysis on the foreground region in the processed foreground image to determine the target connected region corresponding to the seed to be detected;

[0029] Based on the spatial position of the target connected region in the first target image, the target image region is extracted from the corresponding first target image to obtain the seed main image.

[0030] Preferably, the step of performing secondary recognition on the second target image according to a preset second seed image recognition strategy to obtain the target seed recognition result corresponding to the nucleat to be verified includes:

[0031] Obtain the seed type corresponding to the nucleidocyte to be replicated;

[0032] Based on the seed type, the feature category to be extracted corresponding to the seed type is called from the preset feature extraction strategy library, and the target seed recognition strategy corresponding to the seed type is called from the preset seed recognition strategy library.

[0033] Local anomaly region detection is performed on the second target image to obtain the target local anomaly region in the second target image;

[0034] Based on the category of features to be extracted, feature extraction is performed on the target local anomaly region to obtain the target local region feature information;

[0035] Based on the target seed identification strategy, the feature information of the target local region is analyzed, and the target seed identification result is obtained based on the analysis results.

[0036] Preferably, obtaining the seed type corresponding to the nucleidocyte to be replicated includes:

[0037] The second target image is subjected to contour feature extraction processing to obtain the shape contour feature parameters of the nucleus to be processed;

[0038] Based on the aforementioned shape profile feature parameters, the length, width, and aspect ratio of the nucleus to be studied are calculated to obtain the size feature parameters;

[0039] Perform global color statistical analysis on the second target image to obtain the overall color distribution feature parameters;

[0040] Based on the shape contour feature parameters, the size feature parameters, and the overall color distribution feature parameters, the seed type corresponding to the nucleus to be replicated is determined from a preset seed type feature template library.

[0041] Preferably, the step of detecting local anomaly regions in the second target image to obtain the target local anomaly regions in the second target image includes:

[0042] Based on the spatial distribution relationship of pixels in the second target image, the second target image is divided into local regions to obtain multiple local image sub-regions;

[0043] For each of the local image sub-regions, the mean color value, mean gray value, and texture statistical parameters of the pixels within the local image sub-region are calculated to obtain the corresponding local region statistical feature parameters;

[0044] The difference analysis is performed between the statistical feature parameters of each local region and the corresponding statistical feature parameters of the seed in the second target image to obtain the feature deviation degree of each local image sub-region;

[0045] Based on a preset feature deviation threshold, the degree of feature deviation is screened to determine local image sub-regions with significant deviations as candidate sub-regions of abnormal regions.

[0046] The candidate sub-regions of the abnormal region are merged and their boundaries are corrected to obtain the target local abnormal region in the second target image.

[0047] Preferably, the step of analyzing the feature information of the target local region according to the target seed identification strategy, and obtaining the target seed identification result based on the analysis result, includes:

[0048] According to the target seed identification strategy, obtain the feature determination rule parameters corresponding to the feature information of the target local region;

[0049] Based on the feature determination rule parameters, feature matching or threshold judgment processing is performed on each feature parameter in the feature information of the target local area to obtain the local lesion determination result corresponding to each target local abnormal area.

[0050] Based on the distribution location and quantity of each local lesion determination result in the second target image, the local lesion determination results are comprehensively analyzed to generate overall lesion determination information of the nucleidoma to be re-examined.

[0051] Based on the overall lesion determination information, the nucleidomastoids to be reclassified are further classified to obtain the target seed identification result.

[0052] Secondly, the present invention provides an intelligent screening system for diseased seeds based on image recognition, characterized in that it comprises:

[0053] A plurality of identification units are used to identify diseased grains in the seed to be tested according to the image recognition-based seed diseased grain identification method described in the first aspect, and to obtain seed diseased grain identification results;

[0054] The intelligent screening unit is used to classify and screen seeds based on the seed lesion identification results and output the lesion screening results.

[0055] In summary, the beneficial effects of the present invention are as follows:

[0056] This invention provides a seed lesion identification method and intelligent screening system based on image recognition. The method includes: acquiring original images of each seed to be detected under preset illumination conditions; preprocessing each original image to obtain preprocessed first target images; performing preliminary identification on each first target image according to a preset first seed image recognition strategy to obtain an initial seed identification result corresponding to each seed to be detected, wherein the initial seed identification result includes lesions, lesions to be verified, and normal lesions; when the initial seed identification result is a lesion to be verified, extracting the first target image corresponding to the lesion to be verified as a second target image; performing secondary identification on the second target image according to a preset second seed image recognition strategy to obtain a target seed identification result corresponding to the lesion to be verified, wherein the target seed identification result includes lesions and normal lesions; and determining the lesion identification result of each seed to be detected based on the initial seed identification result and the target seed identification result. This invention effectively solves the technical problem in the prior art that it is difficult to simultaneously consider recognition efficiency and recognition accuracy, especially the difficulty in reliably identifying seeds with indistinct lesion features, by constructing a hierarchical and progressive dual-strategy image recognition process. Specifically, the process begins by acquiring the original seed images under preset lighting conditions. A unified preprocessing step eliminates differences in lighting and background, ensuring the stability of subsequent identification. Based on this, a first seed image recognition strategy prioritizing recognition efficiency is introduced. This strategy performs rapid preliminary identification of the first target image, directly classifying lesion particles with obvious lesion characteristics and normal particles with clear characteristics. Only seeds with lesion characteristics falling between the two and whose determination is uncertain are marked as particles requiring further verification, thus avoiding the efficiency degradation caused by high-complexity recognition processing for all samples. Subsequently, a second target image is extracted only for the particles requiring verification, and a second seed image recognition strategy prioritizing recognition accuracy is used for secondary identification. More refined image analysis and judgment rules are used to further determine boundary samples, obtaining reliable target seed identification results. Finally, the preliminary identification results and the secondary identification results are combined to output the final lesion particle identification result. Through this two-stage, on-demand, and refined identification approach, the present invention effectively improves the overall accuracy and stability of lesion particle identification without significantly increasing the overall computational burden, achieving a balance between recognition efficiency and accuracy. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, and these are all within the protection scope of the present invention.

[0058] Figure 1This is a schematic diagram of the overall working process of the image recognition-based seed lesion identification method in Embodiment 1 of the present invention;

[0059] Figure 2 This is a flowchart illustrating the process of performing preliminary identification on each of the first target images in Embodiment 1 of the present invention to obtain the initial seed identification results corresponding to each seed to be detected;

[0060] Figure 3 This is a flowchart illustrating the process of performing secondary recognition on the second target image in Embodiment 1 of the present invention to obtain the target seed recognition result corresponding to the nucleidomastoid to be verified.

[0061] Figure 4 This is a schematic diagram of the intelligent screening system for diseased seeds based on image recognition in Embodiment 2 of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. In the description of the present invention, it should be understood that the terms "center," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, the element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Where there is no conflict, embodiments of the present invention and the various features thereof can be combined with each other, all of which are within the scope of protection of the present invention.

[0063] Example 1

[0064] Please see Figure 1 Embodiment 1 of the present invention discloses a seed lesion identification method based on image recognition, the method comprising:

[0065] Acquire the original images of each seed to be tested under preset lighting conditions;

[0066] Specifically, the raw image refers to seed image data obtained directly during the image acquisition stage without any algorithmic processing. It fully reflects the true appearance of the seed to be detected. By acquiring images under preset lighting conditions, basic constraints can be imposed on the acquisition environment, such as limiting the type, intensity, or direction of the light source, thereby reducing random interference caused by changes in ambient light on image brightness, color representation, and contrast. This provides a stable and consistent input data foundation for subsequent image processing and recognition, making seed images acquired from different batches and locations comparable in imaging quality. In the implementation process, lighting conditions can be controlled through fixed shooting devices, standardized light source components, or enclosed acquisition structures, which helps reduce the impact of the external environment on the expression of lesion features and lays a reliable data foundation for the effective execution of subsequent recognition strategies.

[0067] The original images are preprocessed to obtain the preprocessed first target images.

[0068] Specifically, since directly acquired raw images often contain issues such as background interference, uneven brightness, or scale differences, a standardized preprocessing operation is required before entering the recognition stage. This preprocessing creates image data suitable for recognition and analysis, known as the first target image. By standardizing the image quality and structure, information irrelevant to the seed region is weakened or removed, while visual information relevant to lesion identification is preserved and enhanced. Preprocessing may include brightness and contrast adjustment, separation of background and seed regions, image cropping, and standardization of size or format. The first target image is more regular in spatial structure, imaging quality, and data format, effectively improving the stability and execution efficiency of subsequent image recognition strategies.

[0069] According to the preset first seed image recognition strategy, each of the first target images is initially identified to obtain the initial seed recognition result corresponding to each seed to be detected, wherein the initial seed recognition result includes diseased granules, granules to be re-examined, and normal granules;

[0070] Specifically, after obtaining the first target image, the system performs preliminary analysis and judgment on the image based on a preset first seed image recognition strategy. This first recognition strategy focuses on recognition efficiency and is mainly used to quickly distinguish seeds with obvious abnormal visual features, thereby achieving a preliminary classification of the seeds to be detected. Based on features such as overall brightness, color distribution, or the proportion of abnormal areas in the image, the seed status is judged, and the recognition results are divided into three types: diseased seeds, seeds requiring further verification, and normal seeds. Diseased seeds correspond to seeds with significant pathological features, normal seeds correspond to seeds without obvious abnormalities, and seeds requiring further verification are used to identify seeds with insufficiently clear features that require further detailed analysis. While ensuring basic recognition accuracy, this significantly reduces the number of objects processed by subsequent complex recognition strategies, thereby improving the overall efficiency of the recognition process.

[0071] When the initial seed identification result is a nucleate to be verified, the first target image corresponding to the nucleate to be verified is extracted as the second target image;

[0072] Specifically, after initial identification, some seeds do not fully conform to obvious lesion characteristics in terms of image features, nor can they be stably classified into the normal category. These seeds are marked as granules to be reclassified. The first target image corresponding to the granules to be reclassified still retains complete seed information, so there is no need to re-acquire images. Instead, it is directly extracted from existing image data and used as a separate second target image for subsequent processing. The second target image here is not a new imaging result, but a recombination of the image set from the output results of the initial identification stage according to the category conditions. By extracting and recombining only the images corresponding to the granules to be reclassified, the subsequent more refined identification process can be limited to a smaller number of samples with higher judgment difficulty. This image screening method triggered by the identification result allows the system to avoid repeatedly performing complex analysis on all seeds in the overall processing flow, while retaining sufficient image information for subsequent higher-precision identification.

[0073] According to the preset second seed image recognition strategy, the second target image is re-identified to obtain the target seed recognition result corresponding to the particle to be verified, wherein the target seed recognition result includes diseased particles and normal particles;

[0074] Specifically, the seeds corresponding to the second target image have undergone preliminary screening. Their main characteristic is that the lesion features are not obvious or there are local differences. Therefore, the second seed image recognition strategy adopted at this stage focuses more on detailed information and multi-dimensional feature analysis. This recognition strategy can perform more refined image analysis on the local area of ​​the seed and combine multiple image features for comprehensive judgment to distinguish between seeds with mild lesions, early lesions, or those that appear to be normal. By introducing a recognition strategy with higher judgment accuracy at this stage, seeds that cannot be clearly classified in the preliminary recognition stage can be further distinguished, so that the granules to be verified can be clearly identified as lesions or normal granules. This step plays a role in eliminating uncertainty in the overall process, so that the final recognition result no longer contains ambiguous categories, thereby ensuring the integrity and usability of the lesion granule recognition result.

[0075] Based on the initial seed identification results and the target seed identification results, the diseased grain identification results of each of the seeds to be tested are determined.

[0076] Specifically, after completing two levels of identification, both initial seed identification results and target seed identification results for the granules to be verified are available. For seeds identified as pathological or normal granules in the initial identification stage, their identification results are directly used as the final judgment results; for granules to be verified, the target seed identification results output by the secondary identification are used as the final judgment basis. By integrating the two types of identification results, a complete pathological granule identification result covering all seeds to be tested can be formed. This method based on the integration of hierarchical identification results allows the identification process to retain the high efficiency of the initial identification while compensating for the shortcomings of the initial screening in complex samples through secondary identification, ultimately achieving accurate classification of all seeds to be tested. The resulting pathological granule identification results are consistent and meet the practical needs of subsequent screening, sorting, or quality assessment applications.

[0077] Preferably, the preprocessing of each of the original images to obtain the preprocessed target image includes:

[0078] Brightness compensation processing is performed on each of the original images, and contrast adjustment is performed on the compensated images to obtain each corrected image after brightness and contrast adjustment;

[0079] Specifically, the original image refers to the seed image directly acquired under preset lighting conditions. Its brightness distribution is easily affected by the light source position, seed surface reflectivity, and shooting angle, leading to inconsistencies in brightness even within the same batch of seeds. Brightness compensation of the original image adjusts overly dark or bright areas, resulting in a more balanced brightness distribution in the main seed region. Contrast adjustment further differentiates between different grayscale or color ranges, making seed edge contours and surface features clearer. Brightness compensation and contrast adjustment can be implemented using image processing techniques such as histogram-based grayscale redistribution, adaptive brightness correction, or local contrast enhancement. The corrected image, after these processes, reduces the impact of imaging condition differences on subsequent analysis, maintaining relative consistency in visual features across different seed images and improving the stability of subsequent seed region extraction and recognition results.

[0080] Based on the pixel difference between the seed contour and the background region in each corrected image, seed region extraction processing is performed on each corrected image to obtain various sub-region images;

[0081] Specifically, the corrected image typically contains both the seed body region and the background region, which differ significantly in brightness, color distribution, or texture continuity, especially in the seed outline where abrupt changes in pixel features occur. Based on these pixel differences, analysis of the corrected image can identify continuous regions corresponding to the seed shape and distinguish them from the background region, thus completing the seed region extraction. This process can be implemented using methods such as threshold segmentation, edge detection, region growing, or connected component filtering, while incorporating the continuity of the seed shape to constrain the extraction results. The seed region image obtained through this step primarily contains pixel information related to the seed body, effectively eliminating background interference and allowing subsequent processing to focus on the seed's intrinsic features, helping to reduce the influence of irrelevant regions on recognition.

[0082] Each of the seed region images is cropped and converted in image format to obtain the first target image.

[0083] Specifically, after the seed region images are extracted, their size range, resolution ratio, and data format may vary. This inconsistency can affect the uniform processing of feature information by subsequent image recognition strategies. By cropping the seed region images, the image boundaries can be made to fit the outer contour of the seed body, avoiding the retention of too many invalid blank areas. At the same time, format conversion and size normalization of the images ensure that all seed images are consistent in data structure and size specifications. The above processing can be achieved by fixing the output resolution, unifying the color space, or standardizing the image encoding method. The first target image formed after cropping and format conversion can meet the requirement of consistency of the input image for the initial recognition strategy, allowing the subsequent recognition process to be carried out under the same data conditions, improving recognition efficiency and the comparability of results.

[0084] Preferably, please refer to Figure 2 The step of performing preliminary identification on each of the first target images according to a preset first seed image recognition strategy to obtain the initial seed recognition result corresponding to each seed to be detected includes:

[0085] Foreground and background separation processing is performed on each of the first target images to extract the main body region corresponding to the seed to be detected, and the seed main body image is obtained;

[0086] Specifically, the first target image is the seed region image after its size and format have been standardized, which may still contain background remnants or edge interference information. Foreground and background separation processing refers to distinguishing image regions related to the seed body from non-seed regions based on pixel distribution differences, focusing the analysis on the seed body. This processing is typically based on differences between the seed and background in brightness, color, or texture continuity, combined with region connectivity constraints, to avoid misidentifying scattered noise as seed regions. The seed body image obtained through this step mainly contains the complete seed outline and its surface information, reducing the interference of background factors on subsequent feature analysis and providing a stable and consistent image basis for the extraction of subsequent lesion features.

[0087] Perform grayscale analysis or single color channel analysis on the seed body image to obtain the brightness distribution data or color distribution data of the pixels in the seed body image;

[0088] Specifically, seed lesions often manifest in images as localized brightness anomalies, color shifts, or color unevenness. These features are more easily amplified and quantified in grayscale space or a single color channel. By performing grayscale analysis on the main seed image, color information can be converted into brightness variation relationships to describe the light and dark distribution of different areas on the seed surface. When using single color channel analysis, the focus is on the distribution of a specific color component on the seed surface to capture color changes caused by specific lesions. This analysis process can be achieved by statistically analyzing pixel value distribution and generating brightness or color histograms, allowing the pixel features of the main seed region to be expressed in data form, providing a quantifiable basis for subsequent abnormal area identification.

[0089] Based on a preset brightness threshold or color deviation threshold, and combined with the brightness distribution data or color distribution data, the seed main image is subjected to threshold segmentation processing to extract brightness abnormal regions or color abnormal regions, thereby obtaining a candidate set of abnormal regions.

[0090] Specifically, after obtaining the brightness or color distribution data of the main seed region, pixels can be distinguished by setting a brightness threshold or color deviation threshold, identifying pixel regions that significantly deviate from the normal distribution range. The role of threshold segmentation is to separate potential lesion regions from normal regions, allowing abnormal features to form continuous regions in spatial location. This process can be based on fixed threshold rules or statistical distribution ranges, combined with region connectivity filtering, to avoid misidentifying isolated pixels as abnormal regions. The candidate set of abnormal regions formed through this step provides a clear analytical object for subsequent area ratio calculations, enabling the initial identification stage to quickly locate the range of regions that may contain lesion features.

[0091] The ratio of the pixel area of ​​the abnormal region in the candidate abnormal region set to the total pixel area of ​​the seed main image is calculated to obtain the abnormal region area ratio.

[0092] Specifically, the candidate set of abnormal regions consists of multiple consecutive abnormal regions obtained after threshold segmentation, with each abnormal region corresponding to a set of spatially adjacent abnormal pixels. Pixel area refers to the number of pixels within a region. By counting the pixels in each abnormal region, the area of ​​that abnormal region in the image can be obtained. The total pixel area of ​​the seed body image is obtained by statistically analyzing all valid pixels within the body region. The ratio of the pixel area of ​​the abnormal region to the total pixel area of ​​the seed body image yields a quantitative indicator reflecting the proportion of the abnormal region on the entire seed surface. This ratio objectively characterizes the distribution of lesion features on the seed surface, transforming abnormal regions, which originally only had spatial location significance, into numerical features that can be used for classification.

[0093] The area ratio of the abnormal region is compared with a preset area threshold range. Based on the comparison result, the seed to be detected is initially classified and determined to obtain the initial seed identification result.

[0094] Specifically, the area threshold interval describes the range of abnormal region area proportions corresponding to different identification categories. This interval can be set based on historical sample statistics or empirical data. When the abnormal region area proportion is lower than the first threshold interval, there are fewer abnormal features on the seed surface, and the judgment result tends to be a normal seed. When the abnormal region area proportion is higher than the second threshold interval, the abnormal feature coverage is larger, and the judgment result tends to be a diseased seed. When the abnormal region area proportion falls between the two threshold intervals, the abnormal features are insufficient to make a clear judgment directly, and the seed is marked as a seed to be re-examined. Through this classification method based on area proportion intervals, it is possible to quickly distinguish between obviously normal and obviously diseased seeds without introducing complex calculations, while reasonably retaining seeds with features between the two in the subsequent secondary identification stage, improving the efficiency and stability of the overall identification process.

[0095] Preferably, the step of performing foreground and background separation processing on each of the first target images to extract the main body region corresponding to the seed to be detected, and obtaining the seed main body image, includes:

[0096] Based on the brightness distribution features or color distribution features of pixels in each of the first target images, preliminary threshold segmentation processing is performed on each of the first target images to generate foreground candidate images for characterizing the seed region.

[0097] Specifically, the first target image contains both the seed region and the captured background, which typically exhibit stable differences in brightness or color. For example, the seed region may display a relatively concentrated range of brightness or a specific hue, while the background region may have a more uniform distribution or deviate from this range. Based on this difference, brightness statistics or color channel analysis of the pixels in the image can determine the threshold range used to distinguish between the foreground and background. By performing threshold judgment on each pixel of the image, pixels that meet the conditions are marked as foreground, and the remaining pixels are marked as background, thus obtaining the foreground candidate image. This step initially highlights the seed region in the original image at the pixel level, providing a foundation for subsequent refined processing, while reducing the interference of background complexity on the recognition results.

[0098] Based on the foreground candidate image, morphological processing is performed on the foreground region to obtain a processed foreground image that removes interference from non-seed regions. The morphological processing includes opening operations, closing operations, or noise region removal processing.

[0099] Specifically, after thresholding, foreground candidate images may still contain scattered noise points, holes, or broken edges, which can affect the accurate extraction of the real seed region. By introducing morphological processing techniques, the spatial structure of the foreground region can be optimized. For example, opening operations can be used to remove small, isolated noise regions, closing operations can be used to fill in empty areas inside the seed, or area filtering can be used to directly remove regions that clearly do not conform to the seed size characteristics. After these processing steps, the foreground region becomes more continuous and complete in shape, and closer to the outline of the real seed, which is beneficial for the subsequent stable recognition of the main body region.

[0100] Perform connected component analysis on the foreground region in the processed foreground image to determine the target connected region corresponding to the seed to be detected;

[0101] Specifically, in the processed foreground image, the foreground region typically exists as several independent connected regions, each representing a group of spatially connected foreground pixels. Through connected region analysis, these regions can be individually identified, and their attributes such as area, shape, or spatial location can be statistically analyzed. Considering that seeds in an image typically have a large area, a complete outline, and are located in the main image region, the target connected region that best matches the characteristics of the seed to be detected is selected from multiple connected regions. This process distinguishes the real seed region from the residual non-target foreground regions, ensuring the uniqueness and accuracy of subsequent processing.

[0102] Based on the spatial position of the target connected region in the first target image, the target image region is extracted from the corresponding first target image to obtain the seed main image.

[0103] Specifically, after determining the target connected region, the boundary coordinate information of this region in the first target image can be obtained, including its minimum bounding rectangle or enclosing region. Based on this spatial location information, the corresponding region is directly extracted from the original first target image to obtain an image containing only the seed body. This seed body image retains the true texture, brightness, and color information of the original image while removing irrelevant background, allowing subsequent grayscale analysis, color analysis, and anomaly region extraction to focus on the seed body, thereby improving the accuracy of feature calculation and the reliability of the overall recognition process.

[0104] Preferably, please refer to Figure 3 The step of performing secondary recognition on the second target image according to a preset second seed image recognition strategy to obtain the target seed recognition result corresponding to the nucleat to be verified includes:

[0105] Obtain the seed type corresponding to the nucleidocyte to be replicated;

[0106] Specifically, the seeds to be verified refer to those that cannot be clearly identified as diseased or normal seeds in the initial identification stage, and their appearance characteristics are often greatly affected by varietal differences. Seed type is used to characterize the inherent differences in morphology, color, epidermal texture, etc., between different crops or varieties. For example, corn, wheat, and rice all have significant differences in size, color distribution, and disease manifestations. By retrieving the seed category information pre-recorded in the test task, or combining batch input information, manual annotation information, or data provided by upstream systems, the seed type corresponding to the seeds to be verified is determined. This step enables the subsequent identification process to adopt differentiated analysis paths for different seed types, avoiding the risk of misjudgment caused by uniform standards.

[0107] Based on the seed type, the feature category to be extracted corresponding to the seed type is called from the preset feature extraction strategy library, and the target seed recognition strategy corresponding to the seed type is called from the preset seed recognition strategy library.

[0108] Specifically, different seed types emphasize different features in lesion manifestations; some rely more on color changes, while others rely more on texture disruption or local morphological abnormalities. The feature extraction strategy library stores various pre-defined combinations of feature categories, such as color uniformity features, local gray-scale abrupt changes, texture roughness features, or irregular shape features. The seed recognition strategy library stores multiple sets of judgment rules or model parameters. Based on the determined seed type, matching feature categories and recognition strategies are selected from the two strategy libraries, making the secondary recognition process targeted in feature selection and judgment logic. This seed type-based strategy invocation method helps improve the matching degree between recognition rules and target objects, enhancing the stability of recognition results.

[0109] Local anomaly region detection is performed on the second target image to obtain the target local anomaly region in the second target image;

[0110] Specifically, compared to the image used in the initial identification stage, the second target image typically already shows concentrated seed regions suspected of being abnormal, which may still contain only minor lesion features in localized areas. Local abnormality region detection is used to further locate areas that significantly differ from normal tissue within the main seed region, such as areas of color mutation, texture fragmentation, or areas of abnormally concentrated brightness. Through sliding window analysis, local statistical feature comparison, or region scanning based on threshold changes, each local region in the image is detected one by one, and regions that meet the abnormality judgment criteria are marked as target local abnormal regions. This step narrows the focus of analysis from the overall image to key local regions, laying the foundation for subsequent fine feature extraction and high-precision judgment, while effectively reducing the interference of background or normal regions on the secondary identification results.

[0111] Based on the category of features to be extracted, feature extraction is performed on the target local anomaly region to obtain the target local region feature information;

[0112] Specifically, the target local anomaly region is a concentrated area of ​​suspected lesions located in the second target image. This area often contains subtle features that distinguish it from normal seed tissue. The feature categories to be extracted are used to define specific analytical dimensions. For example, color features focus on pixel tone shifts and color distribution dispersion within the local area; texture features reflect changes in surface roughness; and morphological features characterize the boundary shape or area distribution of the anomaly region. According to the called feature categories, corresponding mathematical statistics or pattern calculations are performed on the pixel data within the target local anomaly region to form multidimensional feature information that can quantitatively describe the characteristics of the anomaly region. This step transforms intuitive image anomalies into structured feature data that can be used for subsequent analysis, which helps improve the stability and repeatability of the identification process.

[0113] Based on the target seed identification strategy, the feature information of the target local region is analyzed, and the target seed identification result is obtained based on the analysis results.

[0114] Specifically, the target local region feature information centrally reflects the key attributes of the suspected lesion area in the grain to be re-identified, while the target seed identification strategy is used to define the judgment logic corresponding to these features within different value ranges. By inputting the extracted feature information into the identification strategy matched with the seed type, a comprehensive analysis of each feature dimension is performed. For example, it is determined whether the color deviation exceeds the normal fluctuation range, whether the texture change shows the distribution of disease characteristics, or whether the size of the abnormal area meets the criteria for lesion identification. Based on the above analysis results, a final identification judgment is made on the grain to be re-identified, classifying it as a lesion grain or a normal grain. This step realizes the transformation from local feature analysis to a clear identification conclusion, making the secondary identification results have higher judgment credibility and effectively making up for the shortcomings of the preliminary identification stage in boundary sample processing.

[0115] Preferably, obtaining the seed type corresponding to the nucleidocyte to be replicated includes:

[0116] The second target image is subjected to contour feature extraction processing to obtain the shape contour feature parameters of the nucleus to be processed;

[0117] Specifically, the second target image corresponds to a single seed image that requires further evaluation after initial screening. It contains complete seed shape information. Contour features are used to describe the overall geometric shape of the seed in the image, such as boundary orientation, contour curvature variation, or the degree of shape regularity. By performing edge detection and contour tracking on the second target image, the boundary line between the seed and the background can be accurately extracted and represented in a parametric form, forming shape contour feature parameters. This step provides a basic geometric reference for subsequent size calculation and type matching, enabling different types of seeds to have a quantifiable basis for differentiation at the morphological level.

[0118] Based on the aforementioned shape profile feature parameters, the length, width, and aspect ratio of the nucleus to be studied are calculated to obtain the size feature parameters;

[0119] Specifically, the shape contour feature parameters can reflect the overall distribution of seed boundaries, and based on this, representative size indicators can be further extracted. By performing geometric analysis on the contour envelope region, the maximum length of the seed along the principal axis and the maximum width perpendicular to it are determined, and the aspect ratio, a parameter reflecting the proportional relationship of seed morphology, is calculated. Different types of seeds usually have relatively stable distribution characteristics in terms of size range and proportional features. This step, through quantitative calculation of contour data, transforms the appearance differences of seeds into clear size feature parameters, which helps to improve the accuracy and stability of subsequent type determination.

[0120] Perform global color statistical analysis on the second target image to obtain the overall color distribution feature parameters;

[0121] Specifically, the overall color distribution characteristics are used to describe the color attributes of seeds at a macroscopic level, distinguishing them from the detailed areas of focus in local anomaly analysis. By statistically analyzing the color information of all pixels in the second target image, the distribution of each color channel can be obtained, such as the color mean, variance, or the proportion of the dominant color, thus forming a set of parameters that can characterize the overall hue of the seed. Since different seed species have inherent color differences in maturity, varietal characteristics, etc., global color statistical analysis can provide stable and representative reference information for seed type determination, while avoiding interference from local lesion areas in type identification.

[0122] Based on the shape contour feature parameters, the size feature parameters, and the overall color distribution feature parameters, the seed type corresponding to the nucleus to be replicated is determined from a preset seed type feature template library.

[0123] Specifically, the seed type feature template library pre-stores typical feature combinations of various seeds in terms of shape, size, and overall color distribution. By matching and analyzing the multiple feature parameters extracted from the kernel to be verified with the feature data in the template library, the similarity in morphological proportions and color attributes can be comprehensively evaluated, and the most suitable seed type can be determined accordingly. This step completes the type determination through multi-feature joint matching, avoiding the risk of misjudgment caused by a single feature, and ensuring that the subsequent feature extraction and recognition strategies are based on a reliable type foundation, thereby improving the targeting and consistency of the secondary recognition process as a whole.

[0124] Preferably, the step of detecting local anomaly regions in the second target image to obtain the target local anomaly regions in the second target image includes:

[0125] Based on the spatial distribution relationship of pixels in the second target image, the second target image is divided into local regions to obtain multiple local image sub-regions;

[0126] Specifically, the second target image corresponds to the complete image of a single seed to be verified, where pixels at different spatial locations often exhibit different distribution characteristics. The purpose of local region segmentation is to divide the entire image into multiple sub-regions with relatively independent statistical characteristics, enabling subsequent anomaly analysis to be performed at a finer spatial scale. In actual processing, the second target image can be partitioned according to a preset grid division method, a sliding window method, or an adaptive segmentation method based on seed morphology, so that each local image sub-region contains a continuous and spatially related set of pixels. Through this segmentation method, local anomaly features are not averaged by overall features, which helps to improve the sensitivity to small lesion areas.

[0127] For each of the local image sub-regions, the mean color value, mean gray value, and texture statistical parameters of the pixels within the local image sub-region are calculated to obtain the corresponding local region statistical feature parameters;

[0128] Specifically, local region statistical features are used to describe the distribution of visual attributes within each image sub-region. The color mean reflects the overall tonal characteristics of the sub-region, the grayscale mean characterizes the brightness level, and texture statistical parameters describe the spatial variation of pixels, such as roughness or uniformity. By statistically analyzing all pixels within a local image sub-region, a stable set of feature parameters can be obtained to characterize the visual state of that region. These statistical features can effectively distinguish normal seed surfaces from localized abnormal areas caused by mold, rot, or insect infestation, providing a reliable data foundation for subsequent anomaly detection.

[0129] The difference analysis is performed between the statistical feature parameters of each local region and the corresponding statistical feature parameters of the seed in the second target image to obtain the feature deviation degree of each local image sub-region;

[0130] Specifically, the overall statistical characteristic parameters of a seed are used to reflect its normal appearance at a macroscopic level. By comparing the statistical characteristic parameters of each local region with the overall statistical characteristics, the degree to which a local region deviates from the overall characteristics in terms of color, brightness, or texture can be quantified. The greater the deviation, the more significant the difference between the local region and the overall appearance of the seed, and the more likely it corresponds to a potential lesion location. This step, by introducing an overall reference benchmark, allows anomaly judgment to be based on adaptive analysis rather than fixed absolute thresholds, thereby enhancing adaptability to seeds of different varieties and appearances.

[0131] Based on a preset feature deviation threshold, the degree of feature deviation is screened to determine local image sub-regions with significant deviations as candidate sub-regions of abnormal regions.

[0132] Specifically, the feature deviation threshold is used to distinguish normal local areas from abnormal areas that may contain lesions. This threshold can be set based on historical sample statistics or empirical rules. By comparing the feature deviation degree of each local image sub-region with the threshold, sub-regions that significantly deviate from the overall characteristics of the seed in terms of color, brightness, or texture can be screened out. These screened local image sub-regions often visually correspond to locations of mold, discoloration, corrosion, or abnormal tissue structures. By introducing this screening step, false positives caused by minor fluctuations in illumination or normal texture changes can be effectively eliminated, allowing subsequent processing to focus on candidate regions that truly possess abnormal features, thereby improving the reliability of anomaly detection results.

[0133] The candidate sub-regions of the abnormal region are merged and their boundaries are corrected to obtain the target local abnormal region in the second target image.

[0134] Specifically, in real-world images, the same lesion area is often divided into multiple adjacent or partially overlapping candidate sub-regions of abnormal regions. Region merging is used to integrate spatially adjacent candidate sub-regions with similar features, forming a coherent and complete abnormal region. Boundary correction is used to smooth and correct the edges of the merged region, removing jagged boundaries caused by segmentation errors or noise, making the abnormal region more closely resemble the spatial morphology of the real lesion area. The target local abnormal region obtained after region merging and boundary correction is improved in both spatial integrity and morphological accuracy, which is beneficial for subsequent feature extraction and recognition based on this region, avoiding interference from the fragmentation of the abnormal region on the recognition results.

[0135] Preferably, the step of analyzing the feature information of the target local region according to the target seed identification strategy, and obtaining the target seed identification result based on the analysis result, includes:

[0136] According to the target seed identification strategy, obtain the feature determination rule parameters corresponding to the feature information of the target local region;

[0137] Specifically, the target seed identification strategy describes the judgment logic adopted for specific seed types and lesion morphologies, containing a set of rule parameters used to interpret and constrain the feature judgment process. These rule parameters can be expressed as weight coefficients, threshold ranges, or combined judgment conditions for different feature dimensions, guiding the selection of subsequent feature analysis methods. By extracting judgment rule parameters from the target seed identification strategy that match the feature information of the current target local region, the analysis process can be kept consistent with specific seed types and their common lesion characteristics, thus avoiding the problem of insufficient applicability of general rules across different seeds. This approach helps establish clear and adjustable judgment criteria for subsequent local lesion determination.

[0138] Based on the feature determination rule parameters, feature matching or threshold judgment processing is performed on each feature parameter in the feature information of the target local area to obtain the local lesion determination result corresponding to each target local abnormal area.

[0139] Specifically, the feature information of a target local region typically consists of multiple feature parameters such as color, texture, or shape, and different features play different roles in lesion identification. By matching each feature parameter with the corresponding feature judgment rule parameters, methods such as threshold judgment, interval judgment, or similarity calculation can be used to determine whether a local abnormal region conforms to lesion characteristics. Each target local abnormal region can obtain an independent local lesion judgment result, which is used to characterize whether the region exhibits obvious lesion characteristics. This step transforms continuous feature values ​​into structured judgment information, making the local analysis results comparable and combinable.

[0140] Based on the distribution location and quantity of each local lesion determination result in the second target image, the local lesion determination results are comprehensively analyzed to generate overall lesion determination information of the nucleidoma to be re-examined.

[0141] Specifically, the lesion assessment results of a single local abnormal area cannot fully reflect the overall health status of the seed; therefore, a comprehensive analysis combining multiple local results is necessary. This step considers not only the number of local lesion assessment results but also their spatial distribution characteristics in the second target image, such as whether they are concentrated in key seed regions, whether they exhibit continuous distribution, or whether they spread over a large area. By summarizing the local lesion assessment results at an overall level, assessment information reflecting the overall lesion severity of the nucleus to be examined can be generated, thereby avoiding over-judgment due to minor local abnormalities and also identifying severe lesions formed by the superposition of multiple mild lesion areas.

[0142] Based on the overall lesion determination information, the nucleidomastoids to be reclassified are further classified to obtain the target seed identification result.

[0143] Specifically, the overall lesion determination information describes the lesion status of the granules to be reclassified in a comprehensive sense. This information can be expressed as the lesion severity level, lesion probability range, or final determination label. By comparing this overall determination information with preset classification rules, a clear secondary classification determination can be made for the granules to be reclassified, thus ultimately classifying them as lesion granules or normal granules. This step completes the closed loop from local feature analysis to overall recognition result output, giving the secondary recognition process a clear determination path and stable output results, which helps improve the system's recognition consistency and engineering usability on complex boundary samples.

[0144] Example 2

[0145] Please see Figure 4 Embodiment 2 of the present invention also provides an intelligent screening system for seed lesions based on image recognition, characterized in that it includes:

[0146] Several identification units are used to identify diseased seeds in the seeds to be tested according to the image recognition-based seed diseased seed identification method described in Example 1, and to obtain seed diseased seed identification results.

[0147] The intelligent screening unit is used to classify and screen seeds based on the seed lesion identification results and output the lesion screening results.

[0148] In summary, the embodiments of the present invention provide a seed lesion identification method and intelligent screening system based on image recognition.

[0149] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0150] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the required tasks. The programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0151] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant locality, and corresponding operation entry points shall be provided for the user to choose to authorize or refuse.

[0152] It should also be noted that the exemplary embodiments mentioned in this invention describe methods or systems based on a series of steps or apparatus. However, this invention is not limited to the order of the steps described above; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0153] The above description is merely a specific embodiment of the present invention. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A method for identifying diseased seeds based on image recognition, characterized in that, The method includes: Acquire the original images of each seed to be tested under preset lighting conditions; The original images are preprocessed to obtain the preprocessed first target images. According to the preset first seed image recognition strategy, each of the first target images is initially identified to obtain the initial seed recognition result corresponding to each seed to be detected, wherein the initial seed recognition result includes diseased granules, granules to be re-examined, and normal granules; When the initial seed identification result is a nucleate to be verified, the first target image corresponding to the nucleate to be verified is extracted as the second target image; According to the preset second seed image recognition strategy, the second target image is re-identified to obtain the target seed recognition result corresponding to the particle to be verified, wherein the target seed recognition result includes diseased particles and normal particles; Based on the initial seed identification results and the target seed identification results, the diseased grain identification results of each of the seeds to be tested are determined; The step of performing secondary recognition on the second target image according to the preset second seed image recognition strategy to obtain the target seed recognition result corresponding to the nucleat to be verified includes: Obtain the seed type corresponding to the nucleidocyte to be replicated; Based on the seed type, the feature category to be extracted corresponding to the seed type is called from the preset feature extraction strategy library, and the target seed recognition strategy corresponding to the seed type is called from the preset seed recognition strategy library. Local anomaly region detection is performed on the second target image to obtain the target local anomaly region in the second target image; Based on the category of features to be extracted, feature extraction is performed on the target local anomaly region to obtain the target local region feature information; Based on the target seed identification strategy, the feature information of the target local region is analyzed, and the target seed identification result is obtained based on the analysis results. The step of performing local anomaly region detection on the second target image to obtain the target local anomaly region in the second target image includes: Based on the spatial distribution relationship of pixels in the second target image, the second target image is divided into local regions to obtain multiple local image sub-regions; For each of the local image sub-regions, the mean color value, mean gray value, and texture statistical parameters of the pixels within the local image sub-region are calculated to obtain the corresponding local region statistical feature parameters; The difference analysis is performed between the statistical feature parameters of each local region and the corresponding statistical feature parameters of the seed in the second target image to obtain the feature deviation degree of each local image sub-region; Based on a preset feature deviation threshold, the degree of feature deviation is screened to determine local image sub-regions with significant deviations as candidate sub-regions of abnormal regions. The candidate sub-regions of the abnormal region are merged and their boundaries are corrected to obtain the target local abnormal region in the second target image.

2. The seed lesion identification method based on image recognition according to claim 1, characterized in that, The step of preprocessing each of the original images to obtain the preprocessed target image includes: Brightness compensation processing is performed on each of the original images, and contrast adjustment is performed on the compensated images to obtain each corrected image after brightness and contrast adjustment; Based on the pixel difference between the seed contour and the background region in each corrected image, seed region extraction processing is performed on each corrected image to obtain various sub-region images; Each of the seed region images is cropped and converted in image format to obtain the first target image.

3. The seed lesion identification method based on image recognition according to claim 1, characterized in that, The step of performing preliminary identification on each of the first target images according to a preset first seed image recognition strategy to obtain the initial seed recognition result corresponding to each seed to be detected includes: Foreground and background separation processing is performed on each of the first target images to extract the main body region corresponding to the seed to be detected, and the seed main body image is obtained; Perform grayscale analysis or single color channel analysis on the seed body image to obtain the brightness distribution data or color distribution data of the pixels in the seed body image; Based on a preset brightness threshold or color deviation threshold, and combined with the brightness distribution data or color distribution data, the seed main image is subjected to threshold segmentation processing to extract brightness abnormal regions or color abnormal regions, thereby obtaining a candidate set of abnormal regions. The ratio of the pixel area of ​​the abnormal region in the candidate abnormal region set to the total pixel area of ​​the seed main image is calculated to obtain the abnormal region area ratio. The area ratio of the abnormal region is compared with a preset area threshold range. Based on the comparison result, the seed to be detected is initially classified and determined to obtain the initial seed identification result.

4. The seed lesion identification method based on image recognition according to claim 3, characterized in that, The step of performing foreground and background separation processing on each of the first target images to extract the main body region corresponding to the seed to be detected, and obtaining the seed main body image, includes: Based on the brightness distribution features or color distribution features of pixels in each of the first target images, preliminary threshold segmentation processing is performed on each of the first target images to generate foreground candidate images for characterizing the seed region. Based on the foreground candidate image, morphological processing is performed on the foreground region to obtain a processed foreground image that removes interference from non-seed regions. The morphological processing includes opening operations, closing operations, or noise region removal processing. Perform connected component analysis on the foreground region in the processed foreground image to determine the target connected region corresponding to the seed to be detected; Based on the spatial position of the target connected region in the first target image, the target image region is extracted from the corresponding first target image to obtain the seed main image.

5. The seed lesion identification method based on image recognition according to claim 1, characterized in that, The step of obtaining the seed type corresponding to the nucleosome to be replicated includes: The second target image is subjected to contour feature extraction processing to obtain the shape contour feature parameters of the nucleus to be processed; Based on the aforementioned shape profile feature parameters, the length, width, and aspect ratio of the nucleus to be studied are calculated to obtain the size feature parameters; Perform global color statistical analysis on the second target image to obtain the overall color distribution feature parameters; Based on the shape contour feature parameters, the size feature parameters, and the overall color distribution feature parameters, the seed type corresponding to the nucleus to be replicated is determined from a preset seed type feature template library.

6. The seed lesion identification method based on image recognition according to claim 1, characterized in that, The step of analyzing the feature information of the local region of the target according to the target seed identification strategy, and obtaining the target seed identification result based on the analysis result, includes: According to the target seed identification strategy, obtain the feature determination rule parameters corresponding to the feature information of the target local region; Based on the feature determination rule parameters, feature matching or threshold judgment processing is performed on each feature parameter in the feature information of the target local area to obtain the local lesion determination result corresponding to each target local abnormal area. Based on the distribution location and quantity of each local lesion determination result in the second target image, the local lesion determination results are comprehensively analyzed to generate overall lesion determination information of the nucleidoma to be re-examined. Based on the overall lesion determination information, the nucleidomastoids to be reclassified are further classified to obtain the target seed identification result.

7. A seed diseased grain intelligent screening system based on image recognition, characterized in that, include: A plurality of identification units are used to identify diseased seeds in the seed to be tested according to any one of claims 1-6, and to obtain the seed diseased seed identification result. The intelligent screening unit is used to classify and screen seeds based on the seed lesion identification results and output the lesion screening results.

Images