Digestive tract focus recognition system for pediatric digestion

Abstract

The invention discloses an alimentary canal focus recognition system for pediatric digestion. The alimentary canal focus recognition system comprises an image acquisition module, an image preprocessing module, a feature extraction module, a focus recognition module, a recognition result output module and a rehabilitation plan making module. The invention belongs to the technical field of pediatric digestion, and particularly relates to a digestive tract focus recognition system for pediatric digestion, which adopts the scheme that a basic structure is constructed to extract different level features and fuse the different level features, a connection rule is calculated and a connection mode is defined, a receptive field is designed and expanded by an expansion convolution, multi-scale features are fused, and a convolution layer structure is designed. The feature extraction effect is improved through jump connection, and comprehensive and hierarchical features are extracted; image features are extracted by using a convolution kernel to obtain a feature map, the state is updated through a reservoir updating unit, the output of an echo convolution network is calculated, feature processing and target recognition are matched by depending on an output weight, the focus condition is comprehensively judged, and the diagnosis efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of pediatric digestion, and in particular is a digestive tract lesion recognition system for pediatric digestion. Background Art

[0002] The digestive tract lesion recognition system is a system that uses medical imaging technology, image processing technology and pattern recognition technology to process and analyze digestive tract-related images, extract feature information, train models to identify lesions, and help doctors more accurately and quickly find digestive tract lesions and improve diagnostic efficiency and accuracy. However, the existing digestive tract lesion recognition system has technical problems such as chaotic feature extraction structure, difficulty in sorting out the connection mode, inability to extract comprehensive and hierarchical features, resulting in inaccurate digestive tract lesion feature extraction and affecting disease diagnosis; and poor recognition ability for dynamic images and images with time association, resulting in the omission of some lesions with features that change over time, which is not conducive to timely adjustment of treatment plans and reduces diagnostic efficiency and accuracy. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a digestive tract lesion recognition system for pediatric digestion. In view of the technical problems that the feature extraction structure is chaotic, it is difficult to sort out the connection mode, and it is impossible to extract comprehensive and hierarchical features, resulting in inaccurate digestive tract lesion feature extraction and affecting the diagnosis of the disease, the basic structure is constructed to extract and fuse features at different levels, the connection rules are calculated to clarify the connection mode, the extended convolution is designed to expand the receptive field and fuse multi-scale features, the convolution layer structure is designed, and the jump connection is used to improve the feature extraction effect; in view of the poor recognition ability of dynamic images and images with time correlation, some lesions with features that change over time are omitted, which is not conducive to timely adjustment of treatment plans and reduces the efficiency and accuracy of diagnosis. The convolution kernel is used to extract image features to obtain a feature map, the state is updated through the reservoir update unit, the output of the echo convolution network is calculated, the feature processing and target recognition are matched based on the output weight, and the lesion situation is comprehensively judged.

[0004] The present invention provides a digestive tract lesion recognition system for pediatric digestion, comprising an image acquisition module, an image preprocessing module, a feature extraction module, a lesion recognition module, a recognition result output module and a rehabilitation plan formulation module;

[0005] The image acquisition module specifically uses an endoscopic camera to collect image data of the inside of the digestive tract in real time;

[0006] The image preprocessing module is specifically used to improve the image quality and obtain the preprocessed digestive tract image data;

[0007] The feature extraction module specifically constructs a basic structure to extract and fuse features at different levels, calculates connection rules to clarify the connection method, designs extended convolution to expand the receptive field and fuse multi-scale features, designs a convolution layer structure, and uses jump connections to improve feature extraction effects;

[0008] The lesion recognition module specifically uses the convolution kernel to extract image features to obtain a feature map, updates the state through the reservoir update unit, calculates the output of the echo convolution network, matches the feature processing and target recognition based on the output weight, and comprehensively judges the lesion situation;

[0009] The recognition result output module specifically provides professional medical advice based on the lesion recognition result and previous medical history;

[0010] The module for formulating a rehabilitation plan specifically considers treatment methods and focuses on dietary adjustments based on the results of digestive tract lesion identification.

[0011] Furthermore, in the image acquisition module, specifically when performing endoscopic examination, the camera on the endoscope collects images of the inside of the digestive tract in real time to obtain digestive tract image data, carefully observe the state of the mucosa, pay attention to the color, texture, smoothness and depressions of the mucosa, observe whether there are ulcers or bleeding, compare with normal mucosa, and preliminarily judge the scope and characteristics of the abnormality.

[0012] Furthermore, in the image preprocessing module, the noise in the digestive tract image data is removed to make the image clearer, the image quality is improved by adjusting the contrast and brightness, the image is sharpened, the edges and details of the image are highlighted, and the preprocessed digestive tract image data is obtained.

[0013] Furthermore, in the feature extraction module, there are provided a basic structure construction unit, a connection rule calculation unit, a dilated convolution design unit and a convolution layer structure design unit, and the feature extraction module includes the following contents:

[0014] Construct the basic structural unit, based on the encoder-decoder of U-Net. In the encoder part, downsampling is performed to extract features at different levels to extract more general features. In the decoder part, upsampling is performed and fused with the corresponding features of the encoder, which helps to restore detailed information and achieve more accurate processing and understanding of the image.

[0015] Calculate the connection rule unit. The encoder and decoder are connected through downsampling and upsampling to calculate the connection rule. The formula used is as follows:

[0016] ;

[0017] In the formula, represents the encoder of the kth layer, D represents the downsampling operation, represents the encoder of the k-1th layer, represents the decoder of the k-1th layer, represents the decoder of the kth layer, and U represents the upsampling operation;

[0018] Design dilated convolution units. Design three dilated convolution modules between different encoder-decoder stages to ensure that the receptive field is expanded without reducing the spatial resolution, better capture feature information of different scales, introduce FPN to connect convolution layers at different stages, extract global multi-scale features, and realize the fusion and utilization of features at different levels.

[0019] The FPN is a deep learning network structure for target detection, which processes feature information of different scales by constructing a feature pyramid;

[0020] Design the convolutional layer structure unit, use a five-layer convolutional layer structure in each encoder-decoder block of the deep learning network, and use jump connections to alleviate the gradient problem to a certain extent. The gradient problem still appears in the overall model. Therefore, gradually reduce the number of convolutional layers and design four different structures, namely five-layer, four-layer, three-layer and two-layer convolutional layers, to explore the connection mode under different structures and the impact on the accuracy of the model, find a more suitable convolutional layer structure to alleviate the gradient problem and improve the feature extraction effect. The formula used is as follows:

[0021] ;

[0022] In the formula, Represents units at different levels in the convolutional layer. By clarifying the units at different levels, we can accurately define the role of each unit in the entire network structure, the calculation method and the relationship with other units. We can deeply understand how units at different positions participate in the process of feature extraction and information transmission. k represents the number of convolutional layers, i is used to index the downsampling layer along the encoder, and j is used to index the convolutional layer of the dense block along the skip path. The range of i and j is [0, 4-k]. H() represents the double convolution operation function, D represents the downsampling operation, and U represents the upsampling operation.

[0023] Furthermore, in the lesion identification module, there are provided a unit for calculating the output of the convolution layer, a unit for performing a flattening operation, a unit for calculating the reservoir update, a unit for calculating the output of the echo convolution network, and a unit for calculating the output weight matrix. The lesion identification module includes the following contents:

[0024] The output unit of the convolution layer is calculated to extract local features of the image. The image data is input into the network and features are extracted through multiple convolution groups of CNN. The parameters of the convolution kernel are randomly initialized and fixed. The convolution kernel matrix is ​​K and the size is 3*3. The output of the convolution layer is calculated. The formula used is as follows:

[0025] ;

[0026] In the formula, represents the final output of the convolution, f() represents the activation function, m represents the number of rows of the convolution layer, n represents the number of columns of the convolution layer, and b represents the bias term added to each element after the convolution operation. represents the weight of the mth row and nth column in the convolution kernel matrix K, and a represents the corresponding element value in the input image matrix;

[0027] The flattening operation unit is used to convert the multi-dimensional feature map into a one-dimensional feature map to solve the problem of inconsistent feature dimensions. After feature extraction, the feature maps with the same number of convolution kernels are obtained. The feature maps are combined into one-dimensional vectors, which are fed into the echo convolution network to calculate the flattened image feature vector. The formula used is as follows:

[0028] ;

[0029] Where v is the flattened image feature vector, and flatten() represents the flattening operation function;

[0030] The computational reservoir update unit is used to dynamically adapt the network to the image data and capture temporal dependencies, which helps to identify lesions more accurately. The flattened image feature vector is divided into multiple time steps and input into the echo convolutional network, represented as u(t), eliminating the complex preprocessing required by the echo convolutional network for image processing tasks. The update equation of the echo convolutional network consists of two parts: updating the storage state and calculating the output. The computational reservoir update equation is as follows:

[0031] ;

[0032] In the formula, represents the reservoir state vector of the w-th width sub-reservoir in the d-th depth sub-reservoir at time t+1, represents the reservoir state vector of the w-th width sub-reservoir in the d-th depth sub-reservoir at time t, represents the input vector at time t+1, tanh represents the hyperbolic tangent function, represents the input connection weight matrix, represents the reservoir neuron connection weight matrix;

[0033] Calculate the output unit of the echo convolutional network, further process and refine the features, the formula used is as follows:

[0034] ;

[0035] In the formula, represents the output of the echo convolutional network at the d-th depth sub-reservoir at time t+1, represents the output weight of the dth layer, Attn represents the self-attention mechanism function, represents the input of the echo convolutional network at the dth depth sub-reservoir at time t+1;

[0036] The output weight matrix unit is calculated, and the output weight matrix is ​​used to better match the feature processing inside the network with the image recognition target. The formula used is as follows:

[0037] ;

[0038] In the formula, represents the final calculated output weight matrix, X represents the set of all reservoir states, that is, the state information related to the features in the network, Y represents the target output matrix, which is the label of the image in image recognition, and T represents the transpose of the matrix.

[0039] Furthermore, in the recognition result output module, the clarity of the lesion boundary, the surface smoothness and whether there are abnormal changes in the surrounding mucosa are described based on the image recognition results. Combined with the past medical history, the results are compared with the previous examination results to give professional medical advice.

[0040] Furthermore, in the rehabilitation plan making module, there are provided a condition assessment unit, a treatment plan making unit and a diet adjustment unit, and the rehabilitation plan making module includes the following contents:

[0041] The condition assessment unit assesses the severity of the patient's condition based on the patient's digestive tract lesion identification results;

[0042] The treatment plan formulation unit considers the patient's overall health, age and underlying diseases, and considers appropriate treatment methods, which include drug therapy, using targeted drugs to relieve symptoms and inhibit the development of lesions, and may also require surgery to remove lesions;

[0043] The dietary adjustment unit focuses on dietary adjustments and recommends that patients eat easily digestible and nutritionally balanced foods, and avoid irritating, high-fat, high-sugar foods that will increase the burden on the digestive tract.

[0044] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0045] (1) In order to solve the technical problems of chaotic feature extraction structure, difficulty in clarifying the connection mode, inability to extract comprehensive and hierarchical features, inaccurate feature extraction of digestive tract lesions, and affecting disease diagnosis, we constructed a basic structure to extract and fuse features at different levels, calculated the connection rules to clarify the connection mode, designed an extended convolution to expand the receptive field and fuse multi-scale features, designed a convolution layer structure, and used jump connections to improve the feature extraction effect;

[0046] (2) In order to solve the technical problem that the recognition ability of dynamic images and images with time correlation is poor, which leads to the omission of some lesions with characteristics that change over time, making it difficult to adjust the treatment plan in time and reducing the efficiency and accuracy of diagnosis, the convolution kernel is used to extract image features to obtain a feature map, and the state is updated through the reservoir update unit. The output of the echo convolution network is calculated, and the feature processing and target recognition are matched based on the output weights to comprehensively judge the lesion situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of a digestive tract lesion identification system for pediatric digestion provided by the present invention;

[0048] Figure 2 is a schematic diagram of the feature extraction module;

[0049] Figure 3 is a schematic diagram of a lesion recognition module;

[0050] Figure 4 Schematic diagram of the modules for developing a rehabilitation plan.

[0051] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0053] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0054] Example 1, see Figure 1 , a digestive tract lesion recognition system for pediatric digestion provided by the present invention includes an image acquisition module, an image preprocessing module, a feature extraction module, a lesion recognition module, a recognition result output module and a rehabilitation plan formulation module;

[0055] The image acquisition module specifically uses an endoscopic camera to collect image data of the inside of the digestive tract in real time;

[0056] The image preprocessing module is specifically used to improve the image quality and obtain the preprocessed digestive tract image data;

[0057] The feature extraction module specifically constructs a basic structure to extract and fuse features at different levels, calculates connection rules to clarify the connection method, designs extended convolution to expand the receptive field and fuse multi-scale features, designs a convolution layer structure, and uses jump connections to improve feature extraction effects;

[0058] The lesion recognition module specifically uses the convolution kernel to extract image features to obtain a feature map, updates the state through the reservoir update unit, calculates the output of the echo convolution network, matches the feature processing and target recognition based on the output weight, and comprehensively judges the lesion situation;

[0059] The recognition result output module specifically provides professional medical advice based on the lesion recognition result and previous medical history;

[0060] The module for formulating a rehabilitation plan specifically considers treatment methods and focuses on dietary adjustments based on the results of digestive tract lesion identification.

[0061] Example 2, see Figure 1 This embodiment is based on the above embodiment. Specifically, in the image acquisition module, when performing endoscopic examination, the camera on the endoscope collects images of the inside of the digestive tract in real time to obtain digestive tract image data, carefully observe the state of the mucosa, pay attention to the color, texture, smoothness and depression of the mucosa, observe whether there is ulcer or bleeding, compare with normal mucosa, and preliminarily judge the scope and characteristics of the abnormality.

[0062] Example 3, see Figure 1This embodiment is based on the above embodiment. In the image preprocessing module, the noise in the digestive tract image data is removed to make the image clearer, the image quality is improved by adjusting the contrast and brightness, the image is sharpened, the edges and details of the image are highlighted, and the preprocessed digestive tract image data is obtained.

[0063] Example 4, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In the feature extraction module, there are provided a basic structure construction unit, a connection rule calculation unit, a dilated convolution design unit and a convolution layer structure design unit. The feature extraction module includes the following contents:

[0064] Construct the basic structural unit, based on the encoder-decoder of U-Net. In the encoder part, downsampling is performed to extract features at different levels to extract more general features. In the decoder part, upsampling is performed and fused with the corresponding features of the encoder, which helps to restore detailed information and achieve more accurate processing and understanding of the image.

[0065] Calculate the connection rule unit. The encoder and decoder are connected through downsampling and upsampling to calculate the connection rule. The formula used is as follows:

[0066] ;

[0067] In the formula, represents the encoder of the kth layer, D represents the downsampling operation, represents the encoder of the k-1th layer, represents the decoder of the k-1th layer, represents the decoder of the kth layer, and U represents the upsampling operation;

[0068] Design dilated convolution units. Design three dilated convolution modules between different encoder-decoder stages to ensure that the receptive field is expanded without reducing the spatial resolution, better capture feature information of different scales, introduce FPN to connect convolution layers at different stages, extract global multi-scale features, and realize the fusion and utilization of features at different levels.

[0069] The FPN is a deep learning network structure for target detection, which processes feature information of different scales by constructing a feature pyramid;

[0070] Design the convolutional layer structure unit, use a five-layer convolutional layer structure in each encoder-decoder block of the deep learning network, and use jump connections to alleviate the gradient problem to a certain extent. The gradient problem still appears in the overall model. Therefore, gradually reduce the number of convolutional layers and design four different structures, namely five-layer, four-layer, three-layer and two-layer convolutional layers, to explore the connection mode under different structures and the impact on the accuracy of the model, find a more suitable convolutional layer structure to alleviate the gradient problem and improve the feature extraction effect. The formula used is as follows:

[0071] ;

[0072] In the formula, Represents units at different levels in the convolutional layer. By clarifying the units at different levels, we can accurately define the role of each unit in the entire network structure, the calculation method and the relationship with other units. We can deeply understand how units at different positions participate in the process of feature extraction and information transmission. k represents the number of convolutional layers, i is used to index the downsampling layer along the encoder, and j is used to index the convolutional layer of the dense block along the skip path. The range of i and j is [0, 4-k]. H() represents the double convolution operation function, D represents the downsampling operation, and U represents the upsampling operation.

[0073] By performing the above operations, the basic structure is constructed to extract and fuse features at different levels, the connection rules are calculated to clarify the connection method, the extended convolution is designed to expand the receptive field and fuse multi-scale features, the convolution layer structure is designed, and the jump connection is used to improve the feature extraction effect. The technical problems of chaotic feature extraction structure, difficulty in clarifying the connection method, and inability to extract comprehensive and hierarchical features, which lead to inaccurate feature extraction of digestive tract lesions and affect the diagnosis of the disease are solved.

[0074] Example 5, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In the lesion recognition module, there are provided a unit for calculating the output of the convolution layer, a unit for performing a flattening operation, a unit for calculating the reservoir update, a unit for calculating the output of the echo convolution network, and a unit for calculating the output weight matrix. The lesion recognition module includes the following contents:

[0075] The output unit of the convolution layer is calculated to extract local features of the image. The image data is input into the network and features are extracted through multiple convolution groups of CNN. The parameters of the convolution kernel are randomly initialized and fixed. The convolution kernel matrix is ​​K and the size is 3*3. The output of the convolution layer is calculated. The formula used is as follows:

[0076] ;

[0077] In the formula, represents the final output of the convolution, f() represents the activation function, m represents the number of rows of the convolution layer, n represents the number of columns of the convolution layer, and b represents the bias term added to each element after the convolution operation. represents the weight of the mth row and nth column in the convolution kernel matrix K, and a represents the corresponding element value in the input image matrix;

[0078] The flattening operation unit is used to convert the multi-dimensional feature map into a one-dimensional feature map to solve the problem of inconsistent feature dimensions. After feature extraction, the feature maps with the same number of convolution kernels are obtained. The feature maps are combined into one-dimensional vectors, which are fed into the echo convolution network to calculate the flattened image feature vector. The formula used is as follows:

[0079] ;

[0080] Where v is the flattened image feature vector, and flatten() represents the flattening operation function;

[0081] The computational reservoir update unit is used to dynamically adapt the network to the image data and capture temporal dependencies, which helps to identify lesions more accurately. The flattened image feature vector is divided into multiple time steps and input into the echo convolutional network, represented as u(t), eliminating the complex preprocessing required by the echo convolutional network for image processing tasks. The update equation of the echo convolutional network consists of two parts: updating the storage state and calculating the output. The computational reservoir update equation is as follows:

[0082] ;

[0083] In the formula, represents the reservoir state vector of the w-th width sub-reservoir in the d-th depth sub-reservoir at time t+1, represents the reservoir state vector of the w-th width sub-reservoir in the d-th depth sub-reservoir at time t, represents the input vector at time t+1, tanh represents the hyperbolic tangent function, represents the input connection weight matrix, represents the reservoir neuron connection weight matrix;

[0084] Calculate the output unit of the echo convolutional network, further process and refine the features, the formula used is as follows:

[0085] ;

[0086] In the formula, represents the output of the echo convolutional network at the d-th depth sub-reservoir at time t+1, represents the output weight of the dth layer, Attn represents the self-attention mechanism function, represents the input of the echo convolutional network at the dth depth sub-reservoir at time t+1;

[0087] The output weight matrix unit is calculated, and the output weight matrix is ​​used to better match the feature processing inside the network with the image recognition target. The formula used is as follows:

[0088] ;

[0089] In the formula, represents the final calculated output weight matrix, X represents the set of all reservoir states, that is, the state information related to the features in the network, Y represents the target output matrix, which is the label of the image in image recognition, and T represents the transpose of the matrix.

[0090] By executing the above operations, the convolution kernel is used to extract image features to obtain a feature map, the state is updated through the reservoir update unit, the output of the echo convolution network is calculated, the feature processing and target recognition are matched by relying on the output weights, and the lesion situation is comprehensively judged. The technical problem that the recognition ability of dynamic images and images with time correlation is poor, resulting in the omission of some lesions with characteristics that change over time, is not conducive to timely adjustment of treatment plans, and reduces the efficiency and accuracy of diagnosis is solved.

[0091] Example 6, see Figure 1 This embodiment is based on the above embodiment. In the recognition result output module, the clarity of the lesion boundary, the surface smoothness, and whether there are abnormal changes in the surrounding mucosa are described according to the image recognition results. Combined with the previous medical history, the results are compared with the previous examination results to give professional medical advice.

[0092] Embodiment 7, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In the rehabilitation plan making module, a condition assessment unit, a treatment plan making unit and a diet adjustment unit are provided. The rehabilitation plan making module includes the following contents:

[0093] The condition assessment unit assesses the severity of the patient's condition based on the patient's digestive tract lesion identification results;

[0094] The treatment plan formulation unit considers the patient's overall health, age and underlying diseases, and considers appropriate treatment methods, which include drug therapy, using targeted drugs to relieve symptoms and inhibit the development of lesions, and may also require surgery to remove lesions;

[0095] The dietary adjustment unit focuses on dietary adjustments and recommends that patients eat easily digestible and nutritionally balanced foods, and avoid irritating, high-fat, high-sugar foods that will increase the burden on the digestive tract.

[0096] It should be noted that, in this article, relational terms such as first and second, etc. are only used 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. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article 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, article or device.

[0097] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0098] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A digestive tract lesion identification system for pediatric digestion, characterized by: The system includes an image acquisition module, an image preprocessing module, a feature extraction module, a lesion recognition module, a recognition result output module and a rehabilitation plan formulation module; The image acquisition module specifically uses an endoscopic camera to collect image data of the inside of the digestive tract in real time; The image preprocessing module is specifically used to improve the image quality and obtain the preprocessed digestive tract image data; The feature extraction module specifically constructs a basic structure to extract and fuse features at different levels, calculates connection rules to clarify the connection method, designs extended convolution to expand the receptive field and fuse multi-scale features, designs a convolution layer structure, and uses jump connections to improve feature extraction effects; The lesion recognition module specifically uses the convolution kernel to extract image features to obtain a feature map, updates the state through the reservoir update unit, calculates the output of the echo convolution network, matches the feature processing and target recognition based on the output weight, and comprehensively judges the lesion situation; The recognition result output module specifically provides professional medical advice based on the lesion recognition result and previous medical history; The module for formulating a rehabilitation plan specifically considers treatment methods and focuses on dietary adjustments based on the results of digestive tract lesion identification.

2. A digestive tract lesion identification system for pediatric digestion according to claim 1, characterized in that: In the feature extraction module, there are provided a basic structure construction unit, a connection rule calculation unit, a dilated convolution design unit and a convolution layer structure design unit. The feature extraction module includes the following contents: Construct the basic structural unit, based on the encoder-decoder of U-Net. In the encoder part, downsampling is performed to extract features at different levels to extract more general features. In the decoder part, upsampling is performed and fused with the corresponding features of the encoder. Calculate the connection rule unit. The encoder and decoder are connected through downsampling and upsampling to calculate the connection rule. The formula used is as follows: ； In the formula, represents the encoder of the kth layer, D represents the downsampling operation, represents the encoder of the k-1th layer, represents the decoder of the k-1th layer, represents the decoder of the kth layer, and U represents the upsampling operation; Design dilated convolution units. Design three dilated convolution modules between different encoder-decoder stages to ensure that the receptive field is expanded without reducing the spatial resolution, better capture feature information of different scales, and introduce FPN to connect convolution layers at different stages to extract global multi-scale features. The FPN is a deep learning network structure for target detection, which processes feature information of different scales by constructing a feature pyramid; Design the convolutional layer structure unit, the formula used is as follows: ； In the formula, Represents units at different levels in the convolutional layer, k represents the number of convolutional layers, i is used to index the downsampling layer along the encoder, j is used to index the convolutional layer of the dense block along the skip path, the range of i and j is [0, 4-k], H() represents the double convolution operation function, D represents the downsampling operation, and U represents the upsampling operation.

3. A digestive tract lesion identification system for pediatric digestion according to claim 1, characterized in that: In the lesion recognition module, there are provided a unit for calculating the output of the convolution layer, a unit for performing a flattening operation, a unit for calculating the reservoir update, a unit for calculating the output of the echo convolution network, and a unit for calculating the output weight matrix. The lesion recognition module includes the following contents: The output unit of the convolutional layer is used to extract local features of the image. The image data is input into the network and the output of the convolutional layer is calculated. The formula used is as follows: ； In the formula, represents the final output of the convolution, f() represents the activation function, m represents the number of rows of the convolution layer, n represents the number of columns of the convolution layer, and b represents the bias term added to each element after the convolution operation. represents the weight of the mth row and nth column in the convolution kernel matrix K, and a represents the corresponding element value in the input image matrix; Perform a flattening operation unit to convert the multi-dimensional feature map into a one-dimensional feature map and calculate the flattened image feature vector. The formula used is as follows: ； Where v is the flattened image feature vector, and flatten() represents the flattening operation function; The computational reservoir update unit is used to dynamically adapt the network to the image data and capture temporal dependencies. The flattened image feature vector is divided into multiple time steps and input into the echo convolutional network to compute the reservoir update equation. The formula used is as follows: ； In the formula, represents the reservoir state vector of the w-th width sub-reservoir in the d-th depth sub-reservoir at time t+1, represents the reservoir state vector of the w-th width sub-reservoir in the d-th depth sub-reservoir at time t, represents the input vector at time t+1, tanh represents the hyperbolic tangent function, represents the input connection weight matrix, represents the reservoir neuron connection weight matrix; Calculate the output unit of the echo convolutional network, further process and refine the features, the formula used is as follows: ； In the formula, represents the output of the echo convolutional network at the d-th depth sub-reservoir at time t+1, represents the output weight of the dth layer, Attn represents the self-attention mechanism function, represents the input of the echo convolutional network at the dth depth sub-reservoir at time t+1; The output weight matrix unit is calculated, and the output weight matrix is ​​used to better match the feature processing inside the network with the image recognition target. The formula used is as follows: ； In the formula, represents the final calculated output weight matrix, X represents the set of all reservoir states, that is, the state information related to the features in the network, Y represents the target output matrix, which is the label of the image in image recognition, and T represents the transpose of the matrix.

4. A digestive tract lesion identification system for pediatric digestion according to claim 1, characterized in that: In the rehabilitation plan making module, there are a condition assessment unit, a treatment plan making unit and a diet adjustment unit. The rehabilitation plan making module includes the following contents: The condition assessment unit assesses the severity of the patient's condition based on the patient's digestive tract lesion identification results; The treatment plan formulation unit comprehensively considers the patient's overall health, age and underlying diseases to formulate a treatment method, which includes drug therapy, using targeted drugs to relieve symptoms and inhibit the development of lesions; The dietary adjustment unit focuses on dietary adjustments and recommends that patients eat easily digestible and nutritionally balanced foods and avoid foods that increase the burden on the digestive tract.

5. The digestive tract lesion identification system for pediatric digestion according to claim 1, characterized in that: In the image acquisition module, specifically when performing endoscopic examination, the camera on the endoscope collects images of the inside of the digestive tract in real time to obtain digestive tract image data, carefully observe the state of the mucosa, pay attention to the color, texture, smoothness and degree of depression of the mucosa, observe whether there are ulcers or bleeding, compare with normal mucosa, and preliminarily judge the scope and characteristics of the abnormality.

6. A digestive tract lesion identification system for pediatric digestion according to claim 1, characterized in that: In the image preprocessing module, the noise in the digestive tract image data is removed to make the image clearer, the image quality is improved by adjusting the contrast and brightness, the image is sharpened, the edges and details of the image are highlighted, and the preprocessed digestive tract image data is obtained.

7. A digestive tract lesion identification system for pediatric digestion according to claim 1, characterized in that: In the recognition result output module, the clarity of the lesion boundary, the surface smoothness, and whether there are abnormal changes in the surrounding mucosa are described based on the image recognition results. Combined with the past medical history, the results are compared with previous examination results to give professional medical advice.

Images