807 results about "Noise removal" patented technology

The invention relates to the technical field of image analysis, in particular to a heat exchanger welding quality visual inspection method and system, and the method comprises the following steps: collecting original image data from a welding image, carrying out the preprocessing of the original image data, including noise removal and brightness adjustment, and obtaining a preprocessed image; according to the method, in the whole process from welding image data collection to defect recognition result generation, preprocessing operation based on high-frequency noise pixel filtering and gray value frequency adjustment is carried out on original image data, interference information in an image is effectively reduced, and the reliability of the image data is improved; meanwhile, continuous iteration deconvolution operation and pixel detail optimization are carried out on the basis of preprocessing, the distinguishing precision and texture detail discernibility of the welding image are improved, the discernibility of the welding defect under different scales is enhanced through multi-scale-level image feature decomposition, and the precision of image feature data is comprehensively improved.

The invention relates to the technical field of industrial automatic quality inspection, and provides a PCBA surface defect detection method and system based on deep learning and a medium, and the method is used for carrying out defect detection on a preset PCBA board. Comprising the following steps: acquiring a surface image of a PCBA board according to a preset multi-angle light source and a high-resolution camera, and performing adaptive illumination compensation and noise removal processing on the surface image to generate a standardized image; performing multi-scale segmentation on the standardized image to obtain image blocks including local details and a global structure; constructing a double-branch deep learning model, wherein the double-branch deep learning model comprises a backbone network, a multi-scale feature fusion module and a defect detection branch; inputting the image blocks into a double-branch deep learning model, and outputting a thermodynamic diagram and probability distribution by the double-branch deep learning model; performing binarization processing on the thermodynamic diagram by using a dynamic threshold segmentation algorithm to generate a defect mask; and outputting a defect detection result of the PCBA board according to the defect mask and the probability distribution, and completing the defect detection of the PCBA board.

The invention relates to the technical field of data analysis, and discloses a data analysis system for multi-joint exercise rehabilitation monitoring. The system collects three-dimensional motion data, myoelectricity activity signals and spatial position data of multiple joints in real time through a multi-mode sensor array. The system comprises a data acquisition module, a preprocessing module, a space-time analysis module and a result output module. The preprocessing module is used for carrying out noise removal, normalization and feature extraction on the original data; the space-time analysis module constructs a monitoring model based on a graph space-time network, and analyzes space-time characteristics of multi-modal data through dynamic joint relation modeling by adopting an intelligent chip. And the result output module generates a visual report including exercise coordination evaluation, injury risk prediction and rehabilitation progress judgment, and feeds back the visual report to the rehabilitation equipment in real time. Besides, the system also has distributed storage and real-time feedback functions, the accuracy and efficiency of rehabilitation monitoring are improved, and the problems of single data processing and inaccurate spatial-temporal feature capture in the prior art are solved.

The invention belongs to the technical field of data exception identification, and particularly relates to a main transformer operation multi-mode data exception identification method, which comprises the following steps: step 1, data acquisition: collecting various data in operation of a main transformer, including current, voltage, temperature, vibration and noise; step 2, data preprocessing: cleaning and sorting the collected data, including noise removal, missing value filling and data smoothing operation; step 3, feature extraction: extracting key features, including statistical features and frequency domain features, from the preprocessed data, and converting the original data into a form for analysis; step 4, modeling and training: utilizing a machine learning or deep learning model to train the features, wherein the model comprises a decision tree and a support vector machine; according to the invention, through a multi-level and multi-dimensional anomaly detection and processing mechanism, the safety and reliability of the operation of the transformer are improved, and the efficiency and cost of operation and maintenance management are optimized.

The invention relates to the technical field of coal mine safety, in particular to a coal mine water disaster prediction system based on a data analysis and machine learning technology, which comprises a multi-source data acquisition module, a dynamic data preprocessing module, a multi-modal feature engineering module, an integrated prediction model construction module and a prediction optimization control module, the multi-source data acquisition module fuses geological and hydrological data, micro-seismic data and equipment working condition data, the dynamic data preprocessing module constructs a noise feature library and realizes noise elimination and data standardization, and the multi-modal feature engineering module extracts dynamic causal feature vectors of a water diversion coefficient change rate and a micro-seismic energy release rate based on convergence cross mapping; the integrated prediction model construction module fuses and outputs a water disaster risk probability value through a meta-learner; and the prediction optimization control module triggers a sampling rate adjustment and disaster response linkage mechanism according to the risk probability value. The method has the advantages of high reliability, high adaptability and timely response, and is suitable for real-time prediction of water disasters in a complex coal mine environment.

The invention discloses an integrated water quality monitoring and processing system for aquaculture, and the system comprises a data collection module which is used for obtaining dissolved oxygen, ammonia nitrogen, pH value, temperature and flow data from a plurality of spatial distributed sensing nodes; the data preprocessing module is used for completing noise removal, correction and time sequence synchronization; the water quality state estimation module is used for outputting real-time multi-point water quality states based on a distributed unscented Kalman filtering algorithm; the space-time fluid modeling module is used for constructing a Lagrange model to extract a flow velocity field and a diffusion coefficient; the scheduling optimization module is used for generating an aeration and dosing scheme by adopting an improved multi-target Pareto optimization algorithm based on a space-time sensitive boundary; the execution module is used for controlling an air blower, an aeration pipe and a dosing pump to complete oxygenation and dosing operations; and the self-adaptive updating module is used for collecting feedback data to adjust filtering gain and optimizing parameters so as to realize closed-loop control of the system.

The invention discloses a transformer fault detection device based on a fuzzy logic algorithm, and the device comprises a data collection module which obtains the operation original data of a transformer in real time through combining the dissolved gas in oil with the temperature, vibration, current and voltage; the data preprocessing module is used for carrying out missing value processing, noise removal and abnormal value detection and processing on the original data; the feature extraction module is used for realizing dynamic feature selection based on data analysis provided by the data preprocessing module; the fault identification module is used for carrying out abnormal waveform judgment on current, voltage, temperature and vibration parameters through a threshold calculation unit and carrying out threshold adjustment based on an optimization algorithm; and the fault detection module triggers the alarm unit or maintains a normal working state according to an identification result of the fault identification module. According to the invention, through monitoring analysis and timely alarm notification, accurate and efficient monitoring of the transformer fault is realized, and stable operation and long-term reliability of the transformer are ensured.

The invention discloses an improved filtering and local direction adaptive rust recognition method for a hull surface rust image, which comprises the following steps of: 1, performing gray processing on the hull surface rust image to obtain a gray image; 2, performing global denoising on the image after gray scale by adopting a wavelet transform and total variation fusion denoising algorithm to obtain a globally denoised image; 3, performing local denoising on the globally denoised image by adopting a self-adaptive Gaussian kernel denoising algorithm to obtain a final denoised image; 4, carrying out the edge extraction of the final denoised image through a direction self-adaptive edge detection algorithm, and obtaining a rust image; according to the method, the improved wavelet transform and total variation fusion denoising algorithm is combined, efficient noise removal is achieved, the size and the standard deviation of the filtering kernel are adjusted in a self-adaptive mode, noise of a local area is effectively processed, detail loss is avoided, and the image quality is improved.

The invention relates to the field of river pollution treatment, and discloses a water quality treatment method, system, equipment and medium based on a remote sensing inversion technology, and the method comprises the following steps: S1, collecting a hyperspectral remote sensing image of a to-be-monitored water body, the hyperspectral remote sensing image comprising chlorophyll concentration, suspended matter concentration and dissolved oxygen; s2, based on the hyperspectral remote sensing image, preprocessing the hyperspectral remote sensing image, including geometric correction, atmospheric correction, noise removal and waveband fusion, to obtain a preprocessed image meeting an inversion precision requirement; and S3, inputting the preprocessed image into a water quality inversion model, and obtaining pixel-level chlorophyll, suspended solids and dissolved oxygen concentration data based on physical radiation transmission. By combining the hyperspectral remote sensing image and the water quality inversion model, large-range, real-time and high-precision water quality monitoring and pollution zoning are realized, model parameters are dynamically updated, and the timeliness and precision of water quality treatment are improved.

The invention discloses a sea surface chlorophyll concentration prediction method and system based on remote sensing and deep learning, and relates to the technical field of marine ecological environment, and the method comprises the steps: inputting the marine chlorophyll a data of a target region into a trained marine chlorophyll concentration prediction model, and obtaining the marine chlorophyll concentration information of the target region; wherein the prediction model is an improved marine chlorophyll a concentration prediction model based on a convolutional long short-term memory network; the periodic feature extraction module is used for extracting historical synchronous related periodic features based on historical synchronous chlorophyll concentration data, and the noise removal and fusion module is used for eliminating noise output by the recent spatial-temporal feature extraction module and the periodic feature extraction module and fusing output results to obtain a marine chlorophyll a concentration data prediction result. According to the invention, the sea surface chlorophyll concentration can be predicted more accurately.

The invention discloses a high-voltage power switch fault diagnosis method based on wavelet threshold correction and noise reduction. Acquiring a voltage signal of the high-voltage power switch by using a sensor, and performing synchronous sampling; a wavelet threshold value correction noise reduction method is adopted to carry out noise reduction processing on the collected signals, wavelet detail coefficients are calculated through multi-scale decomposition, a threshold value is adaptively corrected based on the peak sum ratio, and the noise removal effect is optimized; thirdly, performing normalization processing on the denoised signal, mapping the signal to a polar coordinate system, constructing a two-dimensional Gramer angle field containing an included angle cosine value and amplitude information, and realizing time sequence-space conversion of the signal; and finally, generating two-dimensional image data of the voltage signal of the high-voltage power switch, and providing feature input for subsequent state evaluation and fault detection. According to the method, wavelet transform and two-dimensional feature mapping are combined, noise interference can be effectively reduced, the signal distinguishability and the information retention capacity are improved, and the method is suitable for state monitoring and intelligent diagnosis of a power system.

The invention relates to the field of traffic management, and discloses a traffic situation prediction method based on multi-source heterogeneous data fusion, which comprises the following steps of: firstly, acquiring traffic situation related data of a target area from a plurality of data sources, including traffic flow data, vehicle speed data, video image data, meteorological data and historical traffic statistical data; secondly, preprocessing the acquired traffic situation related data, including data cleaning, normalization or standardization processing, and performing time-space synchronization and matching; wherein the data cleaning comprises noise removal, abnormal value processing and missing value filling; and finally, inputting the preprocessed data into a pre-trained traffic situation prediction model, and outputting the predicted traffic jam degree of the target area. According to the invention, comprehensive analysis is carried out through the traffic-related situation data and the emergency data, and finally, the purpose of improving the prediction comprehensiveness through a multi-source cooperation mechanism is achieved.

The invention relates to the field of intelligent automatic driving, in particular to a multi-vehicle cooperative controllable confrontation test method based on a diffusion model. Comprising the following steps: S1, a diffusion model training stage: training a diffusion model based on real driving data, and learning vehicle behavior distribution through forward noise addition and reverse denoising to obtain fixed model parameters; s2, scene and multi-vehicle space-time modeling: fusing the road map, the lane topology and the vehicle state information, and constructing a space-time scene representation of multi-vehicle interaction as input of diffusion generation; s3, a diffusion generation mechanism based on adversarial guidance; s4, performing partial diffusion control and diversified confrontation generation; and S5, evaluating the authenticity and controllability of the multi-agent confrontation scene. Compared with the prior art, the method has the advantage that the authenticity, controllability and closed-loop consistency of multi-vehicle cooperative confrontation scene generation are obviously improved.

The invention discloses a satellite signal modulation recognition method based on an enhanced multi-scale feature attention network, and the method comprises the steps: obtaining a to-be-recognized satellite signal, inputting the to-be-recognized satellite signal into a trained enhanced multi-scale feature attention network, and obtaining a satellite signal modulation recognition result. The training process of the enhanced multi-scale feature attention network comprises the following steps: acquiring a satellite communication data set, and performing sample expansion to obtain an augmented sample; performing noise removal processing and feature extraction on the augmented sample to obtain denoised data features; inputting the de-noised data features into a global context sensing module, extracting modulation features of different frequencies and time scales, and performing feature fusion on the modulation features and the de-noised data features to obtain fusion features; and performing feature reconstruction and channel recovery on the fusion features, inputting the fusion features into a Softmax classifier, performing classification prediction, calculating classification loss, and optimizing network parameters through the classification loss. Under a complete blind recognition scene, the accuracy of a satellite signal modulation recognition result can be obviously improved.

The invention provides a three-dimensional visualization expression method for a multi-scale geological structure model, including: collecting geological exploration data which includes geophysical information, drilling information, outcrop survey information, seismic information and geological age data; performing interpolation simulation on the geological exploration data to obtain an exploration sample set; preprocessing the exploration sample set to obtain preprocessed data, the preprocessing process including data cleaning, format unification and noise removal; constructing a multi-scale geological structure model according to the preprocessed data based on a three-dimensional modeling technology; performing scale decomposition on the geological structure model using a multi-scale analysis method to form model representations at different scale levels; and converting the model representations into 3D visualization images using a 3D graphics processing technology. The invention can effectively improve the expression ability of geological model and user experience, and has important practical application value for geological exploration and resource assessment.

The invention relates to a method of MR imaging of an object positioned in the examination volume of an MR system (1). It is an object of the invention to provide a deep learning-based denoising approach that overcomes the Rician bias problem. As a solution, the invention proposes a method comprising the following steps: a) subjecting the object to an imaging sequence comprising RF pulses and switched magnetic field gradients, whereby MR signals are generated, b) acquiring the MR signals, c) reconstructing a complex-valued MR image from the acquired MR signals, d) denoising the MR image using a deep learning algorithm that operates on the real and the imaginary parts of the MR image, and c) computing a magnitude MR image from the denoised complex-valued MR image. According to an aspect of the invention, the deep learning algorithm uses a set of denoising models that are trained using different loss functions. In this way, a trade-off between noise removal and preservation of small image details can be controlled. Moreover, the invention relates to an MR system (1) and to a computer program for an MR system (1).

The invention discloses a low-illumination image enhancement method suitable for a complex night operation scene. The method comprises the following steps: preprocessing a collected low-illumination RGB image to obtain a preprocessed image; multi-scale features are extracted from the shallow convolution projection and the multilayer axial converter unit, and cross-layer attention is used for weighted aggregation to form enhanced features; forward noise addition is executed according to noise scheduling to obtain potential representation representing low-illumination noise distribution, and deep semantic representation is extracted in combination with a U-Net encoder embedded in CBAM; in the decoding stage, details are reconstructed through jump connection and a multi-layer converter, meanwhile, reverse denoising of a diffusion model is introduced to gradually remove noise, and an enhanced image with high brightness and low noise is output; in the training process, bidirectional mapping from low illumination to normal illumination is constructed, a weighted target of self-encoding loss and enhanced loss is introduced to carry out joint optimization on a converter and diffusion parameters, and a low-illumination scene is monitored on a general GPU / edge device.

The invention relates to the cross technical field of signal processing and artificial intelligence, in particular to a multi-source noise removal method and system based on DAE, and the method comprises the steps: 1, carrying out the data collection and feature extraction of multi-source noise and pure signals; step 2, constructing a de-noising recognition knowledge base based on feature analysis; step 3, constructing a deep denoising auto-encoder model based on a knowledge base; step 4, hierarchical training and optimization guided by a mixed loss function of the deep denoising model; step 5, denoising processing of a target signal and output evaluation based on a discrimination model; according to the invention, noise data in multiple fields such as electromagnetism, remote sensing and biological signals are integrated, a dynamic mixing strategy and a data enhancement technology are adopted, a training set which highly simulates a real environment is constructed, and a unique cross-scene adaptation module can perform adaptive adjustment according to signal characteristics of different application scenes; the problem that a traditional method is poor in scene adaptability is solved.

The invention discloses an electrocardiosignal preprocessing system and method based on filtering and deep learning, and relates to the field of electrocardiosignal data processing. The multi-stage adaptive filtering module comprises a baseline drift elimination unit, a power frequency interference suppression unit and a myoelectricity noise removal unit, and all the units are connected in sequence to form pipelined parallel processing; the deep learning fusion module is used for performing deep feature extraction and noise classification on the output signal and feeding back a classification result to the multi-stage adaptive filtering module; the signal quality evaluation module carries out quality evaluation on the electrocardiosignals subjected to multi-stage adaptive filtering and deep learning fusion and judges whether the signal quality is qualified or not; a feature enhancement and standardization module; the technical effects of improving the self-adaptability and robustness, improving the calculation efficiency of heart disease classification, the signal fidelity and the diagnosis reliability, and enhancing the signal quality evaluability and the self-adaptive ability are achieved.

The invention relates to the technical field of electrical equipment temperature intelligent on-line monitoring and control, and discloses an electrical equipment temperature intelligent on-line monitoring system, which comprises a data acquisition module used for acquiring temperature, heat flow and environmental data of equipment in real time through a plurality of sensors; the data preprocessing and fusion module is used for carrying out noise removal and standardization processing on the collected data and fusing the multi-source data into a comprehensive data set; and the system identification module is used for updating the thermodynamic model of the equipment according to the real-time data and the historical data. Through cooperative work of the self-adaptive nonlinear optimization control module, the multi-level feedback control module and the energy optimization and scheduling module, optimal balance between temperature control precision and energy consumption is realized, a cooling system can be dynamically adjusted according to real-time temperature of equipment and environment change, energy waste is reduced to the greatest extent, and the service life of the cooling system is prolonged. And the response speed, the energy efficiency and the equipment stability of the system are improved.

The invention provides a power distribution network abnormal state sensing method and system based on data processing, and relates to the technical field of data processing, and the method comprises the steps: obtaining a target function, and setting a differential evolution parameter; the differential evolution parameters comprise population size, iteration times, variation factors and crossover probability; the configuration of differential evolution parameters is iteratively optimized, various operation data are preprocessed, abnormal value and noise removal and normalization processing are included, and processed data are obtained; and extracting characteristic indexes reflecting the running state of the power distribution network from the processed data, wherein the characteristic indexes comprise statistical characteristics, frequency spectrum characteristics and time frequency characteristics of voltage and current. According to the invention, through real-time data acquisition, feature extraction, state evaluation and abnormity identification and positioning, comprehensive monitoring and timely early warning of the operation state of the power distribution network are realized, and the safety and stability of the power distribution network are effectively improved.

The invention discloses an orchard shortest path planning method based on a laser radar SLAM map. The method is used for autonomous navigation of an orchard operation robot. Firstly, orchard three-dimensional point cloud data are collected by using a laser radar SLAM technology, and a high-precision map is constructed. Then carrying out downsampling and noise removal on the point cloud data, extracting ground point cloud and calculating gradient information; thirdly, segmenting the point cloud through a random forest model, distinguishing passing and non-passing areas, extracting the center coordinates of the fruit tree through an improved DBSCAN algorithm, and judging the row direction of the fruit tree through kernel density estimation; then, a cost map containing a gradient cost layer is constructed, and the path length, gradient information and obstacle distribution are comprehensively considered; an improved A * algorithm is adopted to introduce cost map information, a 16 adjacent mode is used to optimize path smoothness, a Bezier curve is used to further carry out local smoothing processing on the path, and a shortest path is generated. The method integrates multiple technologies, is suitable for a complex orchard environment, and has high robustness and applicability.

The invention relates to the technical field of equipment management and maintenance, and discloses a new energy charging station equipment state monitoring and maintenance optimization method, which comprises the following steps: S1, equipment data acquisition: installing various sensors and counters on equipment for acquiring operation data of the equipment in real time, the data comprises voltage, current, temperature, humidity, equipment use duration and charging times; s2, data preprocessing and noise removal: preprocessing the data acquired in the S1 by using a generative adversarial network to remove noise in the data; and S3, equipment health state prediction: predicting the data processed in the S2 based on deep reinforcement learning and a graph neural network, and generating a health prediction model. Through combination of deep reinforcement learning, a generative adversarial network, a block chain and an intelligent contract technology, accurate equipment health prediction, dynamic market pricing, automatic transaction execution and data security guarantee are realized, and the efficiency and security of equipment management and transaction are improved.

The invention discloses an infrared image denoising method and system based on artificial intelligence, and relates to the technical field of image denoising, and the method comprises the steps: collecting infrared image data, extracting frequency domain spatial features, carrying out Gaussian filtering, restoring the features to an image space, determining a high-frequency feature map, extracting image background information, and determining a low-frequency feature map. And splicing the high-frequency feature map through a generator, and carrying out two-dimensional transpose convolution operation by adopting an encoder based on a convolution transpose self-attention mechanism. According to the method, the frequency domain of an infrared image is converted into high-frequency and low-frequency characteristic decomposition, high-frequency characteristics are extracted based on a Gaussian high-pass filter, space structure information such as edges and textures can be effectively reserved, the importance of the high-frequency characteristics can be weighted through an attention mechanism, exploration of noise can be enhanced, and for the background part of the image, the resolution of the image is improved. The attention module can identify an area with excessive brightness fluctuation, and can synchronize transmission of secondary features in a noise removal process through residual features.

The invention provides a point cloud data processing method and device of a laser radar, equipment and a medium, and the method comprises the steps: carrying out the adaptive local neighborhood construction of each query point cloud, and determining a local neighborhood point set of the query point cloud; determining a plurality of geometric features of each query point cloud based on the local neighborhood point set; performing normalization processing and dynamic weight optimization on the plurality of geometric features of each query point cloud, constructing a comprehensive feature of each query point cloud, dividing the plurality of original point cloud data based on the plurality of comprehensive features, and determining a plurality of point cloud categories; and carrying out noise removal processing on the original point cloud data in each point cloud category, determining updated point cloud data, repeatedly carrying out local neighborhood point set determination and point cloud division processing on the updated point cloud data until a convergence condition is met, and outputting the denoised point cloud data. Effectiveness and anti-interference capability of feature fusion are improved, and intelligent identification and denoising of the point cloud and accurate retention of the effective point cloud are realized.

The invention belongs to the technical field of defect detection, and relates to a PCB defect detection method based on image processing, which comprises the following steps: collecting defect-free PCB image data, and making a data set; loss functions are set for the reconstructed subnet and the segmented subnet respectively, the data set is used for training to obtain the trained reconstructed subnet and the segmented subnet, and the trained reconstructed subnet adopts a diffusion model to carry out noise removal on to-be-detected image data subjected to preprocessing and data enhancement step by step according to time steps to obtain a reconstructed image. And decomposing the reconstructed image into a low-frequency component and a high-frequency component through discrete wavelet transform, inputting the low-frequency component into the trained segmentation subnet, comparing the low-frequency component with the to-be-detected image data, and distinguishing a defect part. According to the method, the number of used samples is far smaller than that of other target detection model samples, discrete wavelet transform is utilized, high-frequency components are extracted for back diffusion denoising, the reasoning speed is increased, and the segmentation and defect detection precision is improved through the extracted low-frequency components.

The invention provides a data processing method and system for an aerosol analyzer, and belongs to the technical field of data processing of aerosol analyzers, and the method comprises the following steps: collecting light absorption signals of aerosol particles according to a light source system and a detector, and generating an original electric signal; preprocessing the original electric signal to obtain preprocessed data; extracting an absorption coefficient, a multiband light attenuation characteristic and a mass concentration parameter of the black carbon aerosol from the preprocessed data; analyzing the features based on a machine learning model, outputting a real-time prediction result of the concentration of the black carbon aerosol, and preprocessing the original electric signal to obtain preprocessed data, including the step of noise removal. According to the method, baseline drift is dynamically corrected through a polynomial fitting algorithm, wavelet transform denoising and a machine learning model are combined, the accuracy and real-time performance of black carbon aerosol concentration detection are remarkably improved, and the problems that in a traditional method, manual intervention is much, and the model generalization ability is poor are solved.

The invention provides a as-built drawing multi-element extraction and vectorization method and device, equipment and a storage medium. Relates to the technical field of as-built drawing processing. The method comprises the following steps: segmenting a as-built drawing to obtain binary mask images of a field parcel, a road, a water channel, a building and a wellhead; performing opening and closing operation, morphological smoothing, noise removal and / or hole filling operation on each binary mask image to generate an optimized single-category mask; dividing the optimized single-category masks into three categories of planar elements, linear elements and point-shaped elements according to geometric types of the elements, and respectively adopting a vectorization strategy to generate vector files; and writing the geographic coordinate system and the projection parameters of the as-built drawing into the corresponding vector file to generate an independent layer file with geographic coordinate reference. According to the method, the as-built drawing element extraction precision, the complex element processing capability and the vectorization quality and efficiency are remarkably improved, and the robustness of the model in a complex scene is enhanced.

Apparatuses, systems, and techniques are presented to reduce noise in audio. In at least one embodiment, a sequence of neural networks is used to remove foreground and background noise from audio including a primary audio signal.

The invention provides a high-precision multi-view image recognition detection method based on a CCD camera, and the method comprises the steps: employing a convolutional neural network to extract the noise perception features of each view image through a multi-view image set shot by a CCD camera array; carrying out semantic segmentation by adopting a self-adaptive multi-view semantic segmentation network; constructing final feature representation of the target area based on the enhanced features and the key features; performing view angle consistency matching based on the final feature representation of the target region by adopting similarity calculation; and based on the final target feature map, combining a classification module to identify the defect type, and outputting the type and position information of the defective product. According to the method, the noise removal of the multi-view image is realized through a self-supervised learning method, and the image features under different views are optimized in combination with a self-adaptive semantic segmentation technology, so that the accurate target recognition of the multi-view image is realized.