Alimentary canal image recognition method and system based on state classification

Abstract

The invention provides an alimentary canal image recognition method and system based on state classification, and relates to the technical field of medical image recognition. According to the method, through correction of the space image template, multi-scale feature extraction and weight level division, the accuracy and robustness of alimentary canal image recognition are remarkably improved, diversified image changes can be coped with, the recognition efficiency is improved through the feature extraction and recognition process based on the feature region and the weight level, and more importantly, the recognition efficiency is improved. And the contextual information of the adjacent image frames is utilized to correct the recognition result, so that the accuracy and reliability of the final recognition result are further optimized, efficient application in a complex scene is supported, and powerful technical support is provided for the fields of medical diagnosis, disease monitoring and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image recognition, and in particular to a method and system for digestive tract image recognition based on state classification. Background Art

[0002] In the field of medical imaging, especially in the diagnosis of gastrointestinal diseases, accurate recognition and analysis of gastrointestinal images play a vital role. Due to various factors such as acquisition methods, individual differences among patients, and disease progression, gastrointestinal images are highly diverse and complex. These images often contain rich detailed information, but are also accompanied by unfavorable factors such as noise and artifacts, which poses a huge challenge to accurate image recognition. Traditional image recognition methods often rely on feature extraction at a single scale, making it difficult to fully capture important information about images at different scales. In addition, for multi-level and multi-structural images such as gastrointestinal images, feature extraction at a single scale often leads to information omission or misjudgment, thus affecting recognition accuracy.

[0003] Therefore, it is necessary to provide a digestive tract image recognition method and system based on state classification to solve the above technical problems. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a digestive tract image recognition method and system based on state classification for solving the problems of insufficient image recognition accuracy, insufficient recognition efficiency and misjudgment of image information.

[0005] The present invention provides a digestive tract image recognition method based on state classification, the recognition method comprising the following steps:

[0006] Acquire multiple historical digestive tract images and corresponding recognition results, and perform spatial calibration on the multiple historical digestive tract images using a preset spatial image template;

[0007] According to the recognition results corresponding to the historical digestive tract images, combined with the multi-scale features extracted by the multi-scale analysis method, the spatially corrected historical digestive tract images are identified as multiple feature areas;

[0008] According to the spatial position distribution of each identified feature area at multiple scales and the contribution of each feature area in the recognition result, the weight level of each feature area at different scales is divided, and a weighted space image template is generated;

[0009] Acquire the digestive tract image to be identified and perform preprocessing to obtain the digestive tract image to be identified after preprocessing;

[0010] The preprocessed digestive tract image to be identified is spatially proofread and then input into a weighted spatial image template, the feature regions of the preprocessed digestive tract image to be identified are divided and corresponding multi-scale weight levels are generated;

[0011] According to the divided feature areas and weight levels, features are extracted at different scales and fused into feature sets;

[0012] Input the fused feature set into the pre-trained state recognition model and output the state recognition result;

[0013] The temporal and spatial continuity between adjacent image frames is analyzed, and the state recognition results are further modified using context information to obtain the final recognition results.

[0014] Preferably, the method of acquiring a plurality of historical digestive tract images and corresponding recognition results, and spatially calibrating the plurality of historical digestive tract images using a preset spatial image template, specifically comprises the following steps:

[0015] Collect multiple historical digestive tract image data containing different pathological conditions, and obtain the medical expert recognition results corresponding to each image as label data;

[0016] Design a standardized spatial image template based on the relative positions of key anatomical structures in standard digestive tract images;

[0017] Image processing technology is used to calibrate each historical digestive tract image against a preset spatial image template.

[0018] Preferably, the identification result corresponding to the historical digestive tract image is combined with the multi-scale features extracted by the multi-scale analysis method to identify the spatially corrected historical digestive tract image into multiple feature areas, which specifically includes the following steps:

[0019] Using any one of the Gaussian pyramid and wavelet transform methods in the multi-scale analysis method, multi-scale feature extraction is performed on the spatially corrected historical digestive tract images;

[0020] According to the extracted multi-scale features and the corresponding recognition results, the image segmentation algorithm is used to divide the spatially corrected historical digestive tract images into multiple feature regions.

[0021] Preferably, the method divides the weight levels of each feature region at different scales according to the spatial position distribution of each identified feature region at multiple scales and the contribution of each feature region in the recognition result, and generates a weighted space image template, which specifically includes the following steps:

[0022] Statistical analysis methods are used to analyze the spatial distribution of each feature region at multiple scales and evaluate the contribution of each feature region in the recognition task;

[0023] Based on the evaluation of the contribution of each feature region in the recognition task, a weight level is assigned to each feature region at different scales to obtain the weight level information of each feature region;

[0024] The weight level information of each feature area is integrated into the spatial image template to generate a weighted spatial image template.

[0025] Preferably, the method of spatially correcting the preprocessed digestive tract image to be identified and inputting it into a weighted space image template, dividing the feature areas of the preprocessed digestive tract image to be identified and generating corresponding multi-scale weight levels specifically includes the following steps:

[0026] Performing spatial calibration on the digestive tract image to be identified after preprocessing and the weighted spatial image template, so that the digestive tract image to be identified after preprocessing and the template are spatially consistent;

[0027] The weighted space image template is used to divide the feature areas of the digestive tract image to be identified after preprocessing, and a weight is assigned to each feature area according to the weight level in the weighted space image template.

[0028] Preferably, the extracting features at different scales according to the divided feature regions and weight levels and fusing them into a feature set specifically includes the following steps:

[0029] At different scales, for the divided feature areas, the feature extraction strategy is set according to their weight levels, that is, for areas with high weight levels, detailed and complex features are extracted, and for areas with low weight levels, simple and sparse features are extracted;

[0030] According to the feature extraction strategy, edge detection is used to extract the features of each feature area at each scale;

[0031] The features of each feature area at each scale are normalized, and the serial fusion method is selected to fuse them to obtain the feature set.

[0032] Preferably, the step of inputting the fused feature set into a pre-trained state recognition model and outputting the state recognition result specifically includes the following steps:

[0033] Using the known features and recognition results of historical digestive tract image data as training sets, a state recognition model is obtained through convolutional neural network training;

[0034] The feature set is used as input, and the state recognition model is used to analyze and output the preliminary recognition result of the digestive tract image to be recognized.

[0035] Preferably, the analyzing of the temporal and spatial continuity between adjacent image frames and further correcting the state recognition result using context information to obtain the final recognition result specifically includes the following steps:

[0036] Analyze the temporal and spatial continuity between adjacent image frames in the preliminary recognition results to extract contextual information;

[0037] Use contextual information to evaluate preliminary recognition results;

[0038] According to the result of evaluating the preliminary recognition result, the erroneous information in the preliminary recognition result is marked and corrected accordingly to obtain the final recognition result.

[0039] A digestive tract image recognition system based on state classification, comprising:

[0040] An acquisition and proofreading module, used to acquire a plurality of historical digestive tract images and corresponding recognition results, and to perform spatial proofreading on the plurality of historical digestive tract images using a preset spatial image template;

[0041] An analysis and identification module is used to identify the spatially corrected historical digestive tract image into multiple feature areas based on the recognition results corresponding to the historical digestive tract image and the multi-scale features extracted by the multi-scale analysis method;

[0042] A weight allocation module is used to divide the weight level of each feature area at different scales according to the spatial position distribution of each feature area at multiple scales and the contribution of each feature area in the recognition result, and generate a weighted space image template;

[0043] A preprocessing module is used to collect the digestive tract image to be identified and perform preprocessing to obtain the digestive tract image to be identified after preprocessing;

[0044] A weight division module is used to input the preprocessed digestive tract image to be identified into a weighted spatial image template after spatial calibration, divide the feature area of ​​the preprocessed digestive tract image to be identified and generate corresponding multi-scale weight levels;

[0045] The feature extraction module is used to extract features at different scales according to the divided feature areas and weight levels, and fuse them into feature sets;

[0046] The recognition module is used to input the fused feature set into the pre-trained state recognition model and output the state recognition result;

[0047] The trimming module is used to analyze the temporal and spatial continuity between adjacent image frames and use the context information to further modify the state recognition result to obtain the final recognition result.

[0048] Compared with the related art, the digestive tract image recognition method and system based on state classification provided by the present invention has the following beneficial effects:

[0049] The present invention significantly improves the accuracy and robustness of digestive tract image recognition through the proofreading of spatial image templates, multi-scale feature extraction and weight level division, and is able to cope with diverse image changes. The feature extraction and recognition process based on feature areas and weight levels improves the recognition efficiency. More importantly, the recognition results are corrected using the contextual information of adjacent image frames, which further optimizes the accuracy and reliability of the final recognition results, supports efficient applications in complex scenarios, and provides strong technical support for medical diagnosis, disease monitoring and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flow chart of a digestive tract image recognition method based on state classification according to the present invention;

[0051] Figure 2 The system block diagram of a digestive tract image recognition system based on state classification of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0053] Embodiment 1

[0054] like Figure 1 As shown, a digestive tract image recognition method based on state classification includes the following steps:

[0055] S100, obtaining a plurality of historical digestive tract images and corresponding recognition results, and spatially calibrating the plurality of historical digestive tract images using a preset spatial image template;

[0056] In the specific implementation process, the specific steps of step S100 are:

[0057] S101, collecting a plurality of historical digestive tract image data containing different pathological conditions, and obtaining the medical expert recognition result corresponding to each image as label data;

[0058] Specifically, through hospital cooperation, medical imaging databases, academic conferences and other channels, we widely collect historically known digestive tract image data, conduct preliminary screening on the collected digestive tract image data, remove blurred, incomplete or duplicate images, and obtain the corresponding pathological diagnosis results of each screened image as label data.

[0059] S102, designing a standardized spatial image template according to the relative positions of key anatomical structures in the standard digestive tract image;

[0060] Specifically, one or more representative standard digestive tract images are selected as references. These images should clearly show the key anatomical structures of the digestive tract, such as the esophagus, stomach, duodenum, etc., and accurately mark the boundaries and relative positions of each key anatomical structure. Based on the positioning results of the key anatomical structures, a standardized spatial image template is designed, where the spatial image template contains a clear coordinate system, scale, and anatomical structure markers so that any digestive tract image can be compared and proofread with it.

[0061] S103, using image processing technology to calibrate each historical digestive tract image with a preset spatial image template;

[0062] Specifically, the image processing algorithm extracts key feature points or feature areas in the historical digestive tract image, and then matches the historical digestive tract image with a preset spatial image template through the SIFT matching algorithm. Specifically, the similarity or distance measurement between the historical digestive tract image and the template is calculated to find the best matching position. Then, the image is spatially proofread according to the matching result. If there is a position deviation or scale difference between the historical digestive tract image and the template, the historical digestive tract image is adjusted through affine transformation or perspective transformation to make it consistent with the spatial image template in space and scale.

[0063] S200, according to the recognition result corresponding to the historical digestive tract image, combined with the multi-scale features extracted by the multi-scale analysis method, the spatially corrected historical digestive tract image is identified as a plurality of feature regions;

[0064] In the specific implementation process, the specific steps of step S200 are:

[0065] S201, using any one of Gaussian pyramid and wavelet transform in the multi-scale analysis method to extract multi-scale features from the spatially corrected historical digestive tract image;

[0066] Specifically, according to the specific application scenario and image characteristics, one of the Gaussian pyramid and wavelet transform is selected for multi-scale analysis. In this embodiment, when there are many high-frequency details in the image and these details need to be accurately captured, wavelet transform is selected; while Gaussian pyramid is suitable for the situation where the image resolution needs to be gradually reduced to observe the image structure at different scales.

[0067] In this embodiment, the spatially corrected digestive tract image is used as input, and wavelet transform (such as Haar wavelet, Daubechies wavelet, etc.) is applied to decompose the image into low-frequency (approximation) and high-frequency (detail) parts. The wavelet transform is continued to be applied to the low-frequency part, and this process is repeated until the desired decomposition level is reached. Each layer of decomposition will generate multiple sub-bands corresponding to information of different directions and frequencies, thereby realizing multi-scale feature extraction.

[0068] S202, dividing the spatially corrected historical digestive tract image into a plurality of feature regions using an image segmentation algorithm according to the extracted multi-scale features and the corresponding recognition results;

[0069] Specifically, K-means clustering is selected from the clustering-based segmentation methods, and the extracted multi-scale texture features are used as input. The K-means algorithm is applied to cluster the image pixels into K categories, and each category corresponds to a feature region. The segmentation results are optimized by adjusting the clustering parameters (such as the number of cluster centers K, the number of iterations, etc.) so that the feature regions are more in line with actual needs, so that the historical digestive tract images after spatial correction can be divided into multiple feature regions.

[0070] S300, according to the spatial position distribution of each identified feature region at multiple scales and the contribution of each feature region in the recognition result, the weight level of each feature region at different scales is divided, and a weighted space image template is generated;

[0071] In the specific implementation process, the specific steps of step S300 are:

[0072] S301, using a statistical analysis method to analyze the spatial position distribution of each feature region at multiple scales, and evaluate the contribution of each feature region in the recognition task;

[0073] Specifically, for each feature region, record its spatial position (coordinates, area, shape parameters, etc.) at different scales, then use statistical analysis methods (histogram, scatter plot, heat map, etc.) to analyze the spatial distribution of these feature regions, based on the spatial position distribution statistics, combined with the existing recognition results, i.e., label data, estimate the contribution of each feature region in the recognition task, where the evaluation process is specifically achieved by calculating the correlation between the feature region and the recognition result, and importance indicators (such as information gain, mutual information, etc.)

[0074] S302, based on the evaluation of the contribution of each feature region in the recognition task, assigning a weight level to each feature region at different scales, and obtaining weight level information of each feature region;

[0075] Specifically, a series of weight levels are set according to actual needs. The weight levels set in this embodiment are 1-5, where level 1 represents the lowest weight and level 5 represents the highest weight. Based on the contribution of each feature area evaluated in step S301, corresponding weight levels are assigned to it at different scales. It should be noted that the weight allocation should follow the following principles: feature areas with high contribution should obtain higher weight levels; feature areas that remain stable at different scales should also obtain higher weight levels; and feature areas with low contribution or large changes with scale should obtain lower weight levels.

[0076] S303, integrating the weight level information of each feature area into the spatial image template to generate a weighted spatial image template;

[0077] Specifically, the weight level information of each feature area is encoded into a form that can be represented in the image, such as different colors, brightness, textures, etc. Then, the position of each feature area is marked in the spatial image template and encoded accordingly according to its weight level information.

[0078] S400, collecting the digestive tract image to be identified and preprocessing it to obtain the digestive tract image to be identified after preprocessing;

[0079] S500, spatially calibrating the preprocessed digestive tract image to be identified and inputting it into a weighted spatial image template, dividing the preprocessed digestive tract image to be identified into feature areas and generating corresponding multi-scale weight levels;

[0080] In the specific implementation process, the specific steps of step S500 are:

[0081] S501, spatially calibrating the digestive tract image to be identified after preprocessing with the weighted space image template, so that the digestive tract image to be identified after preprocessing and the template are spatially consistent;

[0082] S502, applying the weighted space image template to divide the digestive tract image to be identified after preprocessing into feature regions, and assigning a weight to each feature region according to the weight level in the weighted space image template.

[0083] S600, extracting features at different scales according to the divided feature regions and weight levels, and fusing them into a feature set;

[0084] In the specific implementation process, the specific steps of step S600 are:

[0085] S601, at different scales, for the divided feature regions, setting feature extraction strategies according to their weight levels, that is, for regions with high weight levels, extracting detailed and complex features, and for regions with low weight levels, extracting simple and sparse features;

[0086] Specifically, according to the weight level, a different feature extraction strategy is formulated for each feature area. In this embodiment, for areas with high weight levels (such as suspected lesion areas), a more detailed and complex feature extraction method is adopted, specifically multi-scale texture analysis and shape descriptors (such as Hu moments and Zernike moments). For areas with low weight levels, a simple and sparse feature extraction method is adopted. In this embodiment, basic color histograms, edge direction histograms, etc. are used to avoid wasting computing resources on unimportant areas.

[0087] S602, according to the feature extraction strategy, edge detection is used to extract the features of each feature area at each scale;

[0088] Specifically, according to the characteristics and requirements of the feature area, a suitable edge detection algorithm is selected, including but not limited to the Canny edge detector, Sobel operator, Prewitt operator, etc., and then the selected edge detection algorithm is applied to the feature area at each scale, and the parameters of the algorithm, specifically the threshold, filter size, etc., are adjusted to meet the feature extraction requirements of areas with different weight levels. In this embodiment, the Canny edge detector is used to extract fine edge information, and the edge density, length, curvature and other features are calculated to describe the shape and contour of the lesion. In the low-weight area, only the Sobel operator is used for simple edge detection to extract basic features such as edge direction and intensity.

[0089] S603, normalizing the features of each feature region at each scale, and selecting a serial fusion method for fusion to obtain a feature set;

[0090] Specifically, the minimum-maximum normalization or Z-score normalization in the normalization method is used to scale the value of each feature to a uniform range, which helps to eliminate the dimensional differences between different features and improve the accuracy and efficiency of subsequent processing. Then, the normalized features are serially concatenated into a long vector in a certain order to obtain a fused feature set.

[0091] Exemplarily, these features are normalized and then concatenated in a certain order, such as high-weight regions first and low-weight regions later, into a feature vector containing 15 elements.

[0092] S700, inputting the fused feature set into a pre-trained state recognition model, and outputting a state recognition result;

[0093] In the specific implementation process, the specific steps of step S700 are:

[0094] S701, using the features and recognition results in the known historical digestive tract image data as a training set, and obtaining a state recognition model through convolutional neural network training;

[0095] Specifically, a large amount of gastrointestinal image data is collected from medical institutions, databases or research projects to ensure that these data contain gastrointestinal images in different states. The collected images are preprocessed, including image resizing, normalization, enhancement (such as rotation, scaling, flipping, etc.) and noise removal. The preprocessed data set is divided into a training set and a validation set. The features and recognition results of the training set are used as input, and a convolutional neural network architecture is selected for training. The prediction results are calculated through forward propagation, and then the loss between the predicted results and the true results is calculated. The weights and biases of the model are updated through back propagation until the stopping condition is met to obtain a trained state recognition model.

[0096] S702: Taking the feature set as input, analyzing and outputting the preliminary recognition result of the digestive tract image to be recognized through the state recognition model.

[0097] S800, analyzing the temporal and spatial continuity between adjacent image frames, and further correcting the state recognition result using context information to obtain a final recognition result;

[0098] In the specific implementation process, the specific steps of step S800 are:

[0099] S801, analyzing the temporal and spatial continuity between adjacent image frames in the preliminary recognition results, and extracting context information;

[0100] Specifically, each frame of the image in the preliminary recognition result is compared with its adjacent frames (previous frame and next frame), and the comparison focuses on the location, size, shape and other features of the lesion area. Then, feature point matching, optical flow method, template matching, etc. are used to calculate the feature change between adjacent frames. According to the results of the inter-frame feature comparison, the temporal continuity of the feature change is analyzed. For example, under normal circumstances, the state change of the digestive tract should be smooth and continuous, rather than sudden jumps. When a frame of the image is initially identified as containing an ulcer, but no ulcer is detected in its adjacent frames, or the location and size of the ulcer have changed significantly and unreasonably, this may be a recognition error. For spatial continuity, the position change of the lesion area (such as ulcer, tumor) in consecutive frames should be continuous and consistent with the structural characteristics of the digestive tract.

[0101] S802, evaluating the preliminary recognition result using context information;

[0102] Specifically, the consistency between the preliminary recognition result and the contextual information is checked. For example, if a frame in the preliminary recognition result is marked as inflammation, but this is not reasonable based on the contextual information (such as no inflammation is detected in adjacent frames), then the result needs to be further evaluated, and a credibility score is assigned to each preliminary recognition result. The scoring is based on the degree of support of the contextual information, the stability of the preliminary recognition result (such as consistency between multiple frames), the significance of the lesion area, etc. Finally, the wrong recognition result is detected based on the contextual information and the credibility score.

[0103] S803, according to the result of evaluating the preliminary recognition result, correspondingly marking and correcting the erroneous information in the preliminary recognition result to obtain a final recognition result;

[0104] Specifically, according to the detected erroneous recognition results, the erroneous information in the preliminary recognition results is marked and corrected accordingly to obtain the final recognition result.

[0105] Embodiment 2

[0106] like Figure 2 As shown, a digestive tract image recognition system based on state classification applied to a digestive tract image recognition method based on state classification specifically includes:

[0107] An acquisition and proofreading module, used to acquire a plurality of historical digestive tract images and corresponding recognition results, and to perform spatial proofreading on the plurality of historical digestive tract images using a preset spatial image template;

[0108] An analysis and identification module is used to identify the spatially corrected historical digestive tract image into multiple feature areas based on the recognition results corresponding to the historical digestive tract image and the multi-scale features extracted by the multi-scale analysis method;

[0109] A weight allocation module is used to divide the weight level of each feature area at different scales according to the spatial position distribution of each feature area at multiple scales and the contribution of each feature area in the recognition result, and generate a weighted space image template;

[0110] A preprocessing module is used to collect the digestive tract image to be identified and perform preprocessing to obtain the digestive tract image to be identified after preprocessing;

[0111] A weight division module is used to input the preprocessed digestive tract image to be identified into a weighted spatial image template after spatial calibration, divide the feature area of ​​the preprocessed digestive tract image to be identified and generate corresponding multi-scale weight levels;

[0112] The feature extraction module is used to extract features at different scales according to the divided feature areas and weight levels, and fuse them into feature sets;

[0113] The recognition module is used to input the fused feature set into the pre-trained state recognition model and output the state recognition result;

[0114] The trimming module is used to analyze the temporal and spatial continuity between adjacent image frames and use the context information to further modify the state recognition result to obtain the final recognition result.

[0115] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0116] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0117] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. A digestive tract image recognition method based on state classification, characterized in that: The identification method comprises the following steps: Acquire multiple historical digestive tract images and corresponding recognition results, and perform spatial calibration on the multiple historical digestive tract images using a preset spatial image template; According to the recognition results corresponding to the historical digestive tract images, combined with the multi-scale features extracted by the multi-scale analysis method, the spatially corrected historical digestive tract images are identified as multiple feature areas; According to the spatial position distribution of each identified feature area at multiple scales and the contribution of each feature area in the recognition result, the weight level of each feature area at different scales is divided, and a weighted space image template is generated; Acquire the digestive tract image to be identified and perform preprocessing to obtain the digestive tract image to be identified after preprocessing; The preprocessed digestive tract image to be identified is spatially proofread and then input into a weighted spatial image template, the feature regions of the preprocessed digestive tract image to be identified are divided and corresponding multi-scale weight levels are generated; According to the divided feature areas and weight levels, features are extracted at different scales and fused into feature sets; Input the fused feature set into the pre-trained state recognition model and output the state recognition result; The temporal and spatial continuity between adjacent image frames is analyzed, and the state recognition results are further modified using context information to obtain the final recognition results.

2. A digestive tract image recognition method based on state classification according to claim 1, characterized in that: The method of obtaining a plurality of historical digestive tract images and corresponding recognition results, and performing spatial calibration on the plurality of historical digestive tract images using a preset spatial image template, specifically comprises the following steps: Collect multiple historical digestive tract image data containing different pathological conditions, and obtain the medical expert recognition results corresponding to each image as label data; Design a standardized spatial image template based on the relative positions of key anatomical structures in standard digestive tract images; Image processing technology is used to calibrate each historical digestive tract image against a preset spatial image template.

3. The digestive tract image recognition method based on state classification according to claim 1, characterized in that: The method of identifying the spatially corrected historical digestive tract image into multiple feature regions based on the recognition results corresponding to the historical digestive tract image and the multi-scale features extracted by the multi-scale analysis method specifically includes the following steps: Using any one of the Gaussian pyramid and wavelet transform methods in the multi-scale analysis method, multi-scale feature extraction is performed on the spatially corrected historical digestive tract images; According to the extracted multi-scale features and the corresponding recognition results, the image segmentation algorithm is used to divide the spatially corrected historical digestive tract images into multiple feature regions.

4. The digestive tract image recognition method based on state classification according to claim 1, characterized in that: According to the spatial position distribution of each identified feature area at multiple scales and the contribution of each feature area in the recognition result, the weight level of each feature area at different scales is divided, and a weighted space image template is generated, which specifically includes the following steps: Statistical analysis methods are used to analyze the spatial distribution of each feature region at multiple scales and evaluate the contribution of each feature region in the recognition task; Based on the evaluation of the contribution of each feature region in the recognition task, a weight level is assigned to each feature region at different scales to obtain the weight level information of each feature region; The weight level information of each feature area is integrated into the spatial image template to generate a weighted spatial image template.

5. The digestive tract image recognition method based on state classification according to claim 1, characterized in that: The method includes the following steps: spatially correcting the preprocessed digestive tract image to be identified and inputting it into a weighted spatial image template, dividing the feature areas of the preprocessed digestive tract image to be identified and generating corresponding multi-scale weight levels. Performing spatial calibration on the digestive tract image to be identified after preprocessing and the weighted spatial image template, so that the digestive tract image to be identified after preprocessing and the template are spatially consistent; The weighted space image template is used to divide the feature areas of the digestive tract image to be identified after preprocessing, and a weight is assigned to each feature area according to the weight level in the weighted space image template.

6. The digestive tract image recognition method based on state classification according to claim 1, characterized in that: The method of extracting features at different scales according to the divided feature regions and weight levels and fusing them into a feature set specifically includes the following steps: At different scales, for the divided feature areas, the feature extraction strategy is set according to their weight levels, that is, for areas with high weight levels, detailed and complex features are extracted, and for areas with low weight levels, simple and sparse features are extracted; According to the feature extraction strategy, edge detection is used to extract the features of each feature area at each scale; The features of each feature area at each scale are normalized, and the serial fusion method is selected to fuse them to obtain the feature set.

7. The digestive tract image recognition method based on state classification according to claim 1, characterized in that: The step of inputting the fused feature set into a pre-trained state recognition model and outputting the state recognition result specifically includes the following steps: Using the known features and recognition results of historical digestive tract image data as training sets, a state recognition model is obtained through convolutional neural network training; The feature set is used as input, and the state recognition model is used to analyze and output the preliminary recognition result of the digestive tract image to be recognized.

8. The digestive tract image recognition method based on state classification according to claim 1, characterized in that: The analysis of the temporal and spatial continuity between adjacent image frames and further correction of the state recognition result using context information to obtain the final recognition result specifically includes the following steps: Analyze the temporal and spatial continuity between adjacent image frames in the preliminary recognition results to extract contextual information; Use contextual information to evaluate preliminary recognition results; According to the result of evaluating the preliminary recognition result, the erroneous information in the preliminary recognition result is marked and corrected accordingly to obtain the final recognition result.

9. A digestive tract image recognition system based on state classification, applied to a digestive tract image recognition method based on state classification as claimed in any one of claims 1 to 8, characterized in that: The identification system comprises: An acquisition and proofreading module, used to acquire a plurality of historical digestive tract images and corresponding recognition results, and to perform spatial proofreading on the plurality of historical digestive tract images using a preset spatial image template; An analysis and identification module is used to identify the spatially corrected historical digestive tract image into multiple feature areas based on the recognition results corresponding to the historical digestive tract image and the multi-scale features extracted by the multi-scale analysis method; A weight allocation module is used to divide the weight level of each feature area at different scales according to the spatial position distribution of each feature area at multiple scales and the contribution of each feature area in the recognition result, and generate a weighted space image template; A preprocessing module is used to collect the digestive tract image to be identified and perform preprocessing to obtain the digestive tract image to be identified after preprocessing; A weight division module is used to input the preprocessed digestive tract image to be identified into a weighted spatial image template after spatial calibration, divide the feature area of ​​the preprocessed digestive tract image to be identified and generate corresponding multi-scale weight levels; The feature extraction module is used to extract features at different scales according to the divided feature areas and weight levels, and fuse them into feature sets; The recognition module is used to input the fused feature set into the pre-trained state recognition model and output the state recognition result; The trimming module is used to analyze the temporal and spatial continuity between adjacent image frames and use the context information to further modify the state recognition result to obtain the final recognition result.

Images