131results about How to "Improve discrimination" patented technology

The invention provides an emotion recognition method based on electroencephalogram feature fusion and a double-stage attention mechanism, and the method comprises the following steps: A, electroencephalogram signal processing: carrying out the preprocessing of an electroencephalogram signal; and B, double-stage attention feature fusion: in each selected frequency band, adopting a double-stage attention mechanism to fuse the electroencephalogram features, and generating fusion features for emotion classification. And C, double-branch feature extraction: performing double-branch 3D convolution processing on the fused features, extracting multi-scale space-spectral time features, and splicing the multi-scale space-spectral time features along a channel dimension to form uniform features. And D, classification and output: inputting the unified features into a classifier, and generating an emotion category prediction result through a flattening layer and a full connection layer. According to the method, the difference entropy, the power spectrum density and the difference entropy asymmetry feature are fused through unified three-dimensional feature representation, a double-stage attention mechanism is introduced, and high-accuracy emotion recognition is achieved.

The invention discloses a two-stage multi-mode bearing fault diagnosis method based on a pre-training large model, and belongs to the technical field of bearing fault diagnosis. The method aims at solving the problems that a traditional method is poor in generalization and poor in robustness under multiple working conditions and small sample conditions. The method comprises the following steps: firstly, constructing a learnable multi-modal Tokens which comprises a multi-scale patch Token, a feature Token and a fault Token, and realizing efficient extraction and fusion of multi-modal features; a time-frequency semantic fusion module is introduced, and comprehensive time-frequency features are output through adaptive frequency coding, time coding and multi-modal fusion; and inputting the multi-modal feature sequence into a pre-training BERT model, and adopting a two-stage training strategy, in the first stage, performing self-supervised pre-training by taking mask signal reconstruction as a target, and in the second stage, performing parameter fine tuning by taking fault classification as a target. According to the method, the diagnosis accuracy and the cross-working-condition generalization ability under the small sample condition can be remarkably improved.

The invention relates to the technical field of molecular breeding, and discloses a molecular marker related to the LDL-C level of porcine serum, a detection reagent and a breeding method. The invention also discloses a primer group for identifying the molecular marker, and the primer group is used for amplifying the following SNP sites: SNP01, which are obtained on the basis of a pig reference genome Sscrofa11.1 and are positioned on 112bp, 008bp and 261bp of a third chromosome, and the genotype is A / G; sNP02 is located at 112bp, 008bp and 354 bp of a third chromosome, and the genotype of SNP02 is A / G; sNP03 is located at 112bp, 008bp and 583bp of the third chromosome, and the genotype of SNP03 is A / G. The primer group for identifying the molecular marker can quickly and effectively identify the LDL-C level of pig serum, and is used for assisting in breeding pig strains with low cholesterol or specific lipid metabolism characteristics.

The invention relates to the technical field of trajectory data mining and identity recognition, in particular to a trajectory user linking method and device based on graph edge weight optimization and a storage medium. The model is composed of a local graph representation learning module fused with grid semantics, a global relation graph representation learning module of adaptive edge weight, a layered space-time attention network and a track user link module. The method comprises the following steps: carrying out gridding processing on an anonymous track, extracting hierarchical semantic embedding of POI categories, and constructing a local space graph and a global relation graph; introducing a semantic consistency coefficient into the local graph to re-calibrate an edge weight, and dynamically modeling an interaction relationship between tracks and between users and tracks in the global graph through a self-adaptive edge weight learning mechanism; fusing local space-time features and global interaction representation through a layered space-time attention network; and finally, the features are projected to the user space for matching, so that the accuracy and robustness of the track user link are remarkably improved.

The application provides application of a substance for detecting neutrophil and / or total bilirubin level in preparation of a kit for predicting early postoperative hemorrhage risk after endoscopic biliary stent implantation, and provides a related prediction model. The application firstly verifies that the substance for detecting neutrophil and / or total bilirubin level can be used for predicting early postoperative hemorrhage risk after endoscopic biliary stent implantation, and based on this, provides a related prediction model, the model has good discrimination ability, and can provide a valuable evaluation tool for preventing hemorrhage of a patient subjected to endoscopic biliary stent implantation.

The application provides a kind of pathological image benign and malignant classification method and device based on mixed supervision and electronic equipment, it is related to digital pathology image analysis and depth learning technical field, including: obtaining first pathological section and second pathological section;From the first pathological section, divide training and verification section, extract strong supervision sample and construct verification set;From the second pathological section, extract tile and calculate entropy value, based on entropy value, screen tile from negative section with strong supervision sample to form initial training set;After training classification model, use the positive probability predicted by model to combine entropy value, screen semi-supervised difficult example negative sample and semi-supervised positive sample from the second pathological section;Merge to form updated training set to continue training, until performance no longer improves, stop.The application uses a small amount of strong supervision label and a large number of weak supervision section, constructs high-quality training set through entropy value screening and confidence learning, significantly reduces the labeling cost and improves the accuracy of classification model.

The present application relates to the technical field of wireless communication security, in particular to a radio frequency fingerprint identification method for wireless node identity authentication. The method solves the problems of weak radio frequency fingerprint characteristics in short frame wireless signals, insufficient discrimination ability of lightweight models, and difficulty in rigid alignment and distillation of heterogeneous teacher-student models. The method obtains training wireless signal samples with identity labels and converts them into time-frequency spectrum samples; inputs the time-frequency spectrum samples into a teacher recognition model and a student recognition model to obtain classification results and global features; projects the teacher global features and the student global features, generates a cross-dimensional correlation matrix according to the projection feature component correlation, and normalizes to obtain alignment-free attention weights; constructs an alignment-free feature distillation loss based on the weights, and jointly trains to obtain a lightweight radio frequency fingerprint identification model; the model is used to output the identity category of the wireless node to be verified during inference, which is used for wireless node access authentication and physical layer identity authentication.

The application discloses a kind of virtual job safety examination methods based on voice response and operation verification, comprising: configuring examination node in virtual job examination scene, and configuring voice question, preset voice response element, preset legal operation set and consistency verification rule for each examination node;After user enters target examination node, corresponding voice question is output and user voice response is received;Voice response is analyzed, and the structured response result associated with the current examination node is generated;User operation behavior is collected within the preset observation window, and operation event sequence is constructed;The consistency verification result is generated by consistency verification of structured response result and operation event sequence.A kind of virtual job safety examination system based on voice response and operation verification is also disclosed.The present application is to solve the problem of voice response and scene operation in existing virtual job examination, such as mutual fragmentation.

The invention belongs to the technical field of deep learning image classification, and provides a cancer tissue pathology image fine-grained classification method based on SMCNet. Comprising the following steps: S1, setting an SMCNet network model; s2, collecting a cancer pathology image fine-grained classification public data set, training a classification network based on the formed SMCNet network model, and storing training weights; s3, performing network performance evaluation on the training result of the SMCNet model by utilizing the evaluation indexes of the accuracy, the classification precision, the recall rate, the F1 score and the AUC score; and S4, in combination with the training weight and a result visualization program, reporting a classification prediction result of the test set, and visualizing the classification prediction result by using a thermodynamic diagram. The SMCNet model adopted by the invention has stronger feature extraction capability and high discrimination degree for fine-grained categories, space-channel features are subjected to DRA attention of feature extraction by using a self-attention module, and the design thought of model lightweight is considered on the basis of realizing an optimized feature extraction function.

The application discloses a kind of packaging box flaw online detection method and device, it is related to online detection technical field.The method steps include: establishing surface response excitation mechanism to form surface response sequence image set;Response consistency analysis is constructed based on the image set by response consistency, to distinguish normal and abnormal area;Abnormal area image is extracted to obtain abnormal surface feature vector, and based on support vector mechanism, build flaw type identification model, input feature vector and output flaw type identification result, combine abnormal area with flaw type identification result, generate packaging box flaw grading detection result.Based on the packaging box flaw grading detection result, the online detection of packaging box flaw is realized.

A dynamic statistics and result analysis system for watershed river and lake shoreline development and utilization comprises a remote sensing data acquisition module, a data preprocessing and classification module, a database construction module, a change trend analysis module, a river and lake directory and association retrieval module and an interaction management module. The data preprocessing and classification module is used for data preprocessing and statistical classification, the database construction unit is used for constructing a system database, the variation trend analysis module is used for variation trend analysis and statistical report generation, and the river and lake directory and association retrieval module is used for directory maintenance and result association. According to the dynamic statistics and result analysis system for watershed river and lake shoreline development and utilization, a remote sensing data statistical classification algorithm based on a neural network is proposed to classify remote sensing data; and a river-lake shoreline development and utilization change trend analysis algorithm based on deep learning is proposed to analyze the change trend.

The application discloses a kind of education resource sequence recommendation method, device and equipment based on three-view structure distillation and storage medium, the method includes: based on student-course interaction log, the learning sequence of time sequence course index is constructed, three types of view hypergraph are constructed according to semantic, path, concept structure multidimensional feature, three-view course representation matrix is generated by structure coding fusion, and the recommended list is output by inputting education resource recommendation model;Education resource recommendation model generates student state vector by the coding of Transformer, and the total course score is obtained by fusing three-view branch score and predicted;The present application predicts the education resource recommendation by three-view collaborative scoring from multiple perspectives, accurately captures course representation, effectively depicts learning path high-order law, improves the accuracy, explainability and generalization ability of education resource recommendation in sparse interaction and cold start scene.

The invention discloses a color high contrast processing algorithm for four-color electronic paper, and relates to the technical field of electronic paper display technology and image processing, and the method comprises the specific steps: firstly, carrying out feature extraction, converting an original RGB image to an HSV space, carrying out the statistics of the pixel proportion of a brightness interval, screening red and yellow pixels, and calculating the feature and the proportion of four colors; a dynamic clustering center is generated; performing weighted clustering mapping to generate an intermediate image; enhancing the saturation and boundary contrast of the adaptive image in a linkage manner; and finally, modular adaptive output: selecting scheme optimization data, converting the scheme optimization data into a 2-bit format, and outputting the two-bit format after verification. According to the method, through multi-dimensional feature extraction and dynamic clustering center generation, accurate color mapping is realized, and the color rendition degree and comfort degree are improved; and through multi-factor linkage enhancement and modular adaptive output design, the display effect and hardware adaptability are considered, the detail expressive force is enhanced, the integration process is simplified, reliable data transmission is guaranteed, and the application adaptation and display quality is improved.

The invention discloses a multi-model paper retrieval method oriented to academic questions and answers, which relates to the technical field of natural language processing and information retrieval, and comprises the following steps: firstly, constructing a unified corpus and a training data set, and respectively encoding by using models in a first model set and a second target model to generate a document vector set; screening difficult negative samples based on the initial retrieval result of the second model, constructing a comparative learning sample, performing fine adjustment on the comparative learning sample, and recoding a corpus; and taking the fine-tuned second target model and the models in the first model set as a model group, respectively performing target query and performing similarity retrieval by adopting each model in the model group to obtain a corresponding original similarity matrix, screening documents based on the original similarity matrix, and generating a first target document list of the target query. According to the method, the accuracy and robustness of academic literature retrieval can be effectively improved.

The application relates to the technical field of medical image analysis, in particular to a classification method and system based on thyroid or breast nodules, which comprises the following steps: acquiring a continuous image frame sequence to locate a calcification region mapping space coordinate, associating adjacent frame homologous segments, extracting a nodule outer edge feature corresponding to a final frame calcification position, analyzing a gray peak value direction connectivity, extracting a texture aggregation boundary feature and overlapping volume curvature and echo distribution analysis to obtain a classification result. In the application, the spatial position and time variation of the calcification region in the continuous image frame are jointly analyzed to obtain a dynamic migration track corresponding to the nodule outer edge feature, a propagation path is constructed in combination with the direction distribution of the internal gray peak value, the boundary contact feature and the texture aggregation form are matched in a unified space, a multi-dimensional feature set with spatial continuity and direction integrity is formed, the analysis ability of the nodule volume, curvature and internal structure difference is improved, and the stability and distinguishability of the classification result are enhanced.

The invention discloses a self-information-length mixed weighted sorting Lempel-Ziv complexity extraction method, which comprises the following steps: acquiring a biomedical signal, preprocessing the biomedical signal, and exporting a preprocessed signal sequence; a C-C differential statistic is constructed by adopting a C-C algorithm, an optimal hysteresis coefficient is determined, an optimal embedding dimension is further determined based on an error nearest neighbor algorithm, and then a signal sequence is converted into a symbolized sequence based on a sorting mode. Counting the distribution probability of each symbol mode in the sequence so as to calculate the self-information value of each symbol mode; traversing the symbolized sequence, identifying all new modes appearing for the first time, calculating the self-information of each new mode based on the self-information of each symbol mode, and constructing a self-information length mixed weighting operator; and all the mixed weighting operators are summed to obtain a weighting count value, and finally the weighting count value is normalized to obtain a final self-information-length mixed weighting ranking Lempel-Ziv complexity value.

This invention discloses a supervised multimodal hash retrieval method and system based on label enhancement, belonging to the field of artificial intelligence and multimedia retrieval technology. The technical problem this invention aims to solve is how to better capture the similarity information between multimodal data points and achieve better performance and accuracy in multimodal retrieval tasks. The technical solution includes: data preprocessing: acquiring and organizing public datasets of image and text modalities, and dividing each public dataset into training, testing, and retrieval datasets; extracting deep features: using a pre-trained network model to extract features from the raw data of the public datasets of image and text modalities respectively, obtaining deep features of the image modality and the text modality; offline training; variable update and optimization; and online query. The system includes a data preprocessing unit, a feature extraction unit, an offline training unit, a variable update and optimization unit, and an online query unit.

The present application provides a method for predicting the efficacy of umbilical cord blood natural killer cell treatment of neuroblastoma, belonging to the technical field of medical prediction and evaluation. The present application solves the clinical need for early and accurate prediction of the efficacy of immunotherapy for neuroblastoma. The method comprises: obtaining peripheral blood test data of a neuroblastoma patient before multiple reinfusion treatment nodes of umbilical cord blood natural killer cell treatment; based on the processed node-level peripheral blood test data, at least one dynamic linkage feature reflecting the dynamic linkage relationship of the indicators in the patient's body is calculated; at least one static feature index is obtained; the dynamic linkage feature and the static feature index are input into a pre-trained efficacy prediction model to obtain the response probability of the patient to umbilical cord blood natural killer cell immunotherapy. The present application accurately predicts the efficacy of umbilical cord blood natural killer cell treatment of neuroblastoma by analyzing the peripheral blood test data of the patient at multiple time points and constructing dynamic linkage features.

The invention discloses a circulating tumor cell physicochemical screen and application thereof. The circulating tumor cell physicochemical screen comprises a body, a reaction tank is arranged in the body, at least two liquid inlets are formed in one end of the reaction tank, a liquid outlet is formed in the other end of the reaction tank, a plurality of reaction areas are arranged between the liquid inlets and the liquid outlet, and a plurality of micro-columns are arranged in each reaction area; along the direction from the liquid inlet to the liquid outlet, the distance between the micro-columns in each reaction area is gradually reduced. The circulating tumor cells are sorted in a mode of combining physical blocking and chemical adsorption, the distance between the microcolumns in each reaction area is gradually reduced, screening of the CTC cells is facilitated, further observation and judgment are also facilitated, compatibility to various coloring agents is good, and the application range is wide.

This invention discloses a method and system for rock mass integrity discrimination based on core image and parameter fusion, comprising the following steps: Step 1: Acquire core images and divide them into multiple image blocks; Step 2: Generate image block structure state vectors; Step 3: Acquire structural parameters and map them into the latent space, perform structural similarity matching discrimination, and generate structural parameter semantic vectors; Step 4: Construct an improved CoFormer model, inputting the image block structure state vectors into the structure state projection branch and the structural parameter semantic vectors into the parameter semantic constraint branch, and outputting a local rock mass integrity score; Step 5: Construct an asymmetric cross-image block inference structure between image blocks; Step 6: Obtain the evolution result of the overall rock mass structure; Step 7: Output the overall integrity level discrimination result of the corresponding rock mass. This invention combines the improved CoFormer model and the asymmetric cross-image block inference structure to achieve intelligent rock mass integrity discrimination.

This invention discloses a project resource recommendation method and system based on knowledge graph subgraph partitioning, relating to the fields of knowledge graph and intelligent recommendation technology. The method includes: fine-grained subgraph partitioning for three target entities: organizations, experts, and technological achievements; independently learning the representations of each subgraph using a graph attention network, and generating deep semantic representations of entities through weighted fusion via an attention mechanism, which are then cached; extracting a list of technical keywords and a requirement summary from the project requirement text using a large language model, and generating requirement text representations through vectorization and multilayer perceptron encoding; based on the relationships in the knowledge graph, using a cross-attention mechanism with the requirement text representation as the query vector, adaptively fusing the deep representations of related target entities to construct a joint representation; finally, calculating the matching degree to generate a recommendation list, and outputting interpretable recommendation reasons based on weighted weights. This invention improves the accuracy, computational efficiency, and interpretability of project resource recommendations.

The invention discloses a binocular stereo matching depth estimation method and system based on hyperspectral reconstruction feature enhancement. The method comprises the following steps: S1, receiving left and right binocular RGB images; s2, performing high-dimensional spectral feature reconstruction on the left binocular RGB image and the right binocular RGB image by using a multi-stage spectral transformation module, and outputting a multi-channel hyperspectral feature map covering a visible light wave band by learning spectral reflectivity priori of a material; s3, adaptive channel selection and feature recombination are carried out on the multi-channel hyperspectral feature map through a learnable spectrum dimension reduction layer, and a left pseudo-color feature map and a right pseudo-color feature map with enhanced material physical attribute differences are generated; and S4, performing geometric feature extraction and parallax iterative optimization based on the left and right pseudo-color feature maps by using a stereo matching module, and outputting a final compact parallax map. According to the method, optical physical priori is learned through model training, so that low-cost RGB hardware can reproduce high-fidelity spectral features, and the problem of depth perception under complex materials and extreme shadows is effectively solved.

The invention belongs to the technical field of power distribution network leakage current detection, and particularly relates to a CNN and FiLM-based leakage current type intelligent identification method. Comprising the following steps: S1, collecting real-time operation data of a typical power supply area, obtaining leakage current waveforms and related environment characteristic parameters under different working conditions, and constructing an original leakage current sample data set; s2, a CNN-based leakage current classification model is constructed, a FiLM condition modulation module is introduced, a multi-task learning framework is used at the tail of the network, a main task is leakage current type identification, and an auxiliary task is grounding system discrimination; s3, using a weighted cross entropy loss function to alleviate a class sample imbalance problem; an OneCycleLR dynamic learning rate scheduling strategy is introduced; s4, evaluating the performance of the model on the test set, and using the accuracy and the confusion matrix as evaluation indexes; according to the method, high-precision identification of multiple types of faults such as single-phase grounding, arc type electric leakage and direct current system electric leakage is realized, different grounding systems can be adapted, and the accuracy and robustness of system diagnosis are improved.

The invention provides a method and a system for fraud detection by using a contrast graph neural network based on intra-class substructure differentiation, and solves the technical problem of low detection precision caused by heterogeneity and class imbalance of existing similar subclusters. The method comprises the following steps: acquiring original data, and carrying out manual annotation to obtain annotated data; modeling the marked data into a graph structure, inputting the graph structure into a comparison graph neural network model based on substructure differentiation for training, and obtaining a trained comparison graph neural network model after reaching a preset training round; and reasoning the original data modeled as a graph structure by using the trained comparison graph neural network model to obtain a classification result. The method can be widely applied to the technical field of fraud detection.

This invention relates to the field of functional engineered wood products, specifically to an engineered wood product with ultra-long room-temperature phosphorescence characteristics. The engineered wood product includes bonded units and an adhesive layer located between adjacent bonded units. The adhesive layer is formed by an adhesive that, upon curing, can generate room-temperature phosphorescence. Under external light excitation, the adhesive layer can generate ultra-long room-temperature phosphorescence emission, and the emission signal includes at least one of the following: emission color, emission intensity, emission spectrum, delayed emission image, and / or phosphorescence lifetime. The bonded units are selected from at least one of the following: decorative paper, thin wood veneer, engineered wood veneer, wood veneer, wood / bamboo strips, wood / bamboo shavings, wood / bamboo fiber, straw, and biomass fragments. By adjusting the adhesive system, interface structure, and hot-pressing process parameters, the resulting engineered wood product can exhibit different room-temperature phosphorescence characteristics. This invention endows the board with stable room-temperature phosphorescence function without changing the basic manufacturing process of existing engineered wood products, possessing good process compatibility and application prospects. The room-temperature phosphorescence signal can construct material-level optical coding information, realizing the identification and anti-counterfeiting traceability of the engineered wood product.

The invention provides an intelligent fault diagnosis method and system for a krypton and xenon extraction process and a storage medium, and the method specifically comprises the steps: obtaining multivariable time series data of the krypton and xenon extraction process, and inputting the multivariable time series data into a multi-scale residual dense cavity convolutional network to extract a high-dimensional depth feature sequence; the network uses cavity convolution with different expansion rates in different residual blocks, and fuses multi-scale time sequence features through dense connection; weighting the feature sequence by using a channel attention and time self-attention mechanism to obtain a time weighted feature; meanwhile, on the basis of covariance modeling between feature channels, feature dimension attention scores are calculated, and variable correlation weighted features are obtained; performing first outer product fusion on the time weighted feature and the variable correlation weighted feature, performing second outer product fusion on the original depth feature and the time weighted feature, and splicing the two fusion results to generate a fault representation vector; and inputting the fault representation vector into a classifier, and outputting a fault type diagnosis result.

The invention belongs to the technical field of biological information processing, and particularly relates to a method and system for constructing a risk prediction model from Kawasaki disease to huge coronary artery tumor. The invention provides a Kawasaki disease MGCAA risk prediction model construction method and system, and the method comprises the steps: building a prediction model based on six conventional clinical variables through employing a random forest (RF ranger) algorithm; an SHAP explanation mechanism is introduced, so that the contribution of each variable to a prediction result can be visually displayed in global and individual levels, and the transparency and clinical interpretability of the model are improved; through external verification and intercept-only recalibration, the reliability and the applicability of the model in different groups of people can be ensured; according to the method, an online webpage tool (Shiny App) is deployed, and a doctor can input clinical data of a patient in real time and immediately obtain individualized risk prediction and explanation, so that clinical early recognition of an MGCAA high-risk child patient is realized, and formulation of an individualized treatment scheme and early intervention measures is assisted.

This invention discloses a fault diagnosis method for wind turbine turntable bearings based on WDCNN-Transformer, belonging to the field of wind power equipment condition monitoring and intelligent fault diagnosis technology. This invention proposes a novel fault diagnosis method by combining dual-branch time-frequency domain parallel feature extraction with the serial structure of WDCNN-Transformer and time-frequency domain feature fusion. This method effectively improves the accuracy and robustness of wind turbine turntable bearing fault diagnosis. It utilizes both time and frequency domain branches simultaneously, automatically extracting features using deep networks to ensure comprehensive mining of signal information from multiple perspectives. The concatenation of WDCNN and Transformer improves the utilization efficiency of temporal and spatial features. Feature fusion fully leverages the complementary advantages of time-frequency information, enhancing the discriminative power of fault features.

The invention provides a self-adaptive ultrasonic detection and defect identification method for a composite material, and belongs to the technical field of ultrasonic detection.The self-adaptive ultrasonic detection and defect identification method for the composite material comprises the specific steps that firstly, preparation is conducted before detection; 2, collecting full-matrix data of the one-way plate; 3, calculating a core elastic constant of the composite board; 4, deducing the velocity distribution of each layer group of the composite board; 5, collecting full-matrix data of the composite board; step 6, adaptively correcting imaging delay; and 7, performing full-focusing imaging and judgment on the defects of the composite board. According to the method, defect subtle features which are difficult to recognize by human eyes in the full-focus imaging picture can be automatically mined through the deep learning model, the problems of missing detection and false detection caused by subjective judgment deviation are effectively avoided, rapid and accurate classification and positioning of various defects can be realized through the deep learning model, the detection period is greatly shortened, and the detection efficiency is improved. And reliable technical support is provided for high efficiency and precision of ultrasonic detection of the composite material.

The application belongs to the technical field of tunnel lining detection, and specifically discloses a ground penetrating radar high signal-to-noise ratio image imaging method, which comprises the following steps: constructing a training data set of pairs of noisy ground penetrating radar B-scan images and clean images; building an improved U-Net denoising network model, embedding a CBAM attention module in the encoder and the decoder, and enhancing effective signal features and suppressing noise through channel attention and spatial attention; using the training data set to train the model, optimizing network parameters with clean images as expected output; inputting the image to be processed into the trained model, and outputting a high signal-to-noise ratio image. The application can effectively separate disease signals in a strong noise interference background under single disease scene and combined disease scene conditions, and does not appear signal distortion, amplitude attenuation and edge blur, and the denoised image is almost consistent with the original clean image, thereby providing a strong basis for subsequent intelligent disease recognition.