58 results about "Gabor filter" patented technology

A method is provided for improving image classification accuracy in few-shot learning scenarios, where only a limited number of training examples are available. The method combines the use of Gabor filters and convolutional neural networks (CNNs) to extract detailed texture and orientation features from images. These features are then enhanced through global average pooling, aggregated into comprehensive feature vectors, and refined using an attention mechanism that identifies and emphasizes the most relevant features for classification. Masks generated from this attention process selectively enhance critical features, which, after optional re-encoding, are used to train a classifier via a metric learning approach. This method aims to increase feature separability and classification performance, facilitating more accurate classification of new images with minimal training data.

The invention relates to the technical field of image data processing, in particular to a forge piece surface defect detection method and system based on machine vision. The method comprises the steps of obtaining a gray image of a to-be-detected forging surface; determining a boundary significance weight; determining a path consistency weight; screening the boundary significance weight and the path consistency weight to obtain a final weight; and carrying out local adaptive threshold segmentation on the final feature map to obtain a binary image, and carrying out defect identification on the surface of the forge piece to be detected based on a connected region in the binary image. According to the method, the boundary significance weight and the path consistency weight are constructed and are respectively used for accurately positioning defect edges and verifying structure continuity, texture interference is effectively inhibited, and false alarms are reduced; through adaptive Gabor filtering, the problem of a response blind area of a traditional method is solved, finally two weights are fused to modulate filtering response, and the accuracy of forging surface defect detection is improved.

The invention belongs to the technical field of image processing, and particularly relates to a screw fastening quality evaluation method and system based on image features, and the method comprises the steps: collecting a screw fastening image, and obtaining a complex response diagram through a Log-Gabor filter; obtaining a phase congruency diagram according to each complex response diagram; acquiring gradient directions of pixel points in the phase consistency graph, and constructing an accumulator graph according to the gradient directions and offset points in a preset radius range; according to salient points in the accumulator graph and local signal-to-noise ratios of the salient points, screw positioning points are determined through two-dimensional quadratic polynomial function fitting; the torque value of the screw is obtained according to the screw positioning point driving torque measuring device, and whether the screw is fastened or not is judged. According to the method, the screw is positioned by using phase consistency in the frequency domain, so that the interference of specular reflection and extreme shadow on positioning is overcome, the accuracy of screw positioning is improved, and screw fastening quality evaluation is assisted.

The invention relates to a bridge structure modal parameter identification method, which is based on a bridge structure tiny vibration video acquired by a camera. VMD (variational mode decomposition), PBVM (phase-based video motion amplification), Gabor filtering-based phase extraction method GBP (Gabor filtering), an improved clustering algorithm ICA (independent clustering algorithm) and an integrated covariance-driven random subspace recognition (SSI-Cov) and FDD (frequency domain decomposition) algorithm are fused to construct a set of complete non-contact structural modal parameter recognition system. According to the method, effective decoupling of multiple vibration modes of a bridge structure, frequency domain feature extraction of small-amplitude vibration and intelligent and automatic modal parameter extraction are achieved under the conditions that the target frequency bandwidth does not need to be known in advance, environmental noise interference exists and no obvious feature target exists, and the method belongs to the technical field of bridge structure health monitoring.

The invention discloses a space-time image velocity measurement method and system based on multi-direction Gabor filtering, and the method comprises the steps: carrying out the filtering processing of a space-time image through Gabor filters in multiple directions, so as to enhance the texture features in multiple directions, carrying out the image quality evaluation of a Gabor filtering image in each direction, and obtaining the velocity measurement of the space-time image. The filtering direction with the best textural feature enhancement effect is screened out; a clustering point set with the highest comprehensive evaluation score is screened out through a two-layer evaluation screening mode of clustering quality evaluation and line segment quality evaluation, straight line fitting is carried out based on the clustering point set, an optimal flow velocity line can be obtained, and the river surface flow velocity is obtained through calculation; according to the method, a three-layer evaluation system composed of filtering direction evaluation, clustering quality evaluation and line segment quality evaluation is combined, a high-quality, correct and effective flow velocity line can be accurately extracted from a space-time image, and the accuracy and robustness of a flow velocity measurement result are greatly improved.

The invention belongs to the technical field of image analysis and processing, and relates to an automobile circuit board solder paste printing quality detection method and system. The method comprises the steps of constructing a Laplacian pyramid for a target solder paste area image, extracting multi-level geometric and position features, performing weighting based on local phase consistency to obtain a first fusion feature, and comparing the first fusion feature with a first fusion feature of a historical qualified sample to obtain a second fusion feature; calculating a time sequence drift correction characteristic capable of correcting the drift in the production process; constructing a neighborhood feature map by combining the first fusion features of the target region and the spatial neighborhood of the target region, and extracting spatial morphological features by using an orthogonal Gabor filter; and performing vector splicing on the first fusion feature, the time sequence drift correction feature and the spatial form feature to generate a comprehensive combined feature, and inputting the comprehensive combined feature into a predetermined classification model to obtain a printing quality detection result. According to the invention, the detection rate of the solder paste area defect on the automobile circuit board can be improved.

The invention discloses a micro-scratch filtering method and device in wafer defect detection, equipment and a storage medium, and the method comprises the steps: extracting the frequency spectrum characteristics of a wafer image, carrying out the direction characteristic analysis based on the frequency spectrum characteristics, and carrying out the calculation to obtain the center frequency of a target region suspected to have micro-scratches; constructing a Gabor filter according to the center frequency, and reinforcing the target area by using the Gabor filter; respectively carrying out gray integration along the direction vector and the vertical direction vector of the linear defect of the strengthened target area, and calculating the strength attribute of the target area; and when the strength attribute is higher than a preset threshold value, marking the defect of the target area as a micro-scratch and filtering out the micro-scratch. According to the method, the micro scratches in the image can be accurately positioned and filtered out, and the influence of the micro scratches on the accuracy of wafer defect detection is prevented.

The invention discloses a sandstone slice image component identification method based on a GL-SLIC algorithm and semi-supervised learning. The method comprises the following steps: firstly, acquiring a sandstone slice image amplified by 200 times under single polarization, converting an RGB image into a CIE-Lab color space, and extracting a GLBP texture feature vector; then a GL-SLIC superpixel segmentation algorithm is constructed, and adaptive segmentation based on texture and color features is realized in combination with a Gabor filter and a local binary pattern; then implementing a region merging algorithm to generate complete mineral particles and pore fragments; finally, a classifier based on VGG16 and a discriminator based on ResNet18 are constructed, a semi-supervised self-training framework is adopted, a model is initialized by using about 6% of manual labeling samples, a high-confidence-coefficient pseudo-label extension training data set is generated through iteration, and automatic recognition of sandstone components such as quartz, pores, kaolinite, rock debris and a matrix is achieved. According to the method, the recognition accuracy of 96.3% on a test set is achieved, compared with a traditional method, the segmentation precision and the recognition accuracy are remarkably improved, the data labeling cost is greatly reduced, and an efficient technical scheme is provided for automatic analysis of geological images.

The invention belongs to the technical field of image data processing, and particularly relates to a large shaft forging blank center forging method and system based on machine vision, and the method comprises the steps: carrying out the preprocessing of a high-temperature steel ingot image; carrying out convolution on the image by adopting a logarithm-Gabor filter bank to obtain complex response and local direction energy; for each pixel point, obtaining the structural significance of the pixel point based on the sum of the energy in the real part, the imaginary part and the local direction of the complex response; obtaining the direction certainty based on the local direction energy and the direction vector; obtaining a final voting weight by combining the structure significance and the direction certainty; accumulative voting is carried out based on the weight, the geometric axis of the steel ingot is calculated, and an anvil of the forging press is aligned with the geometric axis of the steel ingot. According to the method, weighted voting is carried out by combining the authenticity and directivity of the pixel points, strong interference such as water vapor and oxide skin is effectively restrained, and stable and accurate positioning of the steel ingot axis under the complex working condition is achieved.

The present application relates to the technical field of image data processing, and more particularly to a forging surface defect detection method and system based on machine vision. The method comprises: acquiring a grayscale image of a forging surface to be detected; determining a boundary saliency weight; determining a path consistency weight; screening the boundary saliency weight and the path consistency weight to obtain a final weight; performing local adaptive threshold segmentation on the final feature map to obtain a binary image, and based on the connected regions in the binary image, performing defect recognition on the forging surface to be detected. By constructing the boundary saliency weight and the path consistency weight, the present application is respectively used for accurately positioning the defect edge and verifying the structural continuity, effectively suppresses the texture interference and reduces false alarms; by adaptive Gabor filtering, the response blind area problem of the traditional method is solved, and finally the two weights are fused to modulate the filtering response, thereby improving the accuracy of the forging surface defect detection.

The present application relates to the technical field of coast water edge line detection, and is an island water edge line detection method based on phase consistency random walk, comprising: converting a single polarization SAR image into a phase consistency image by using a two-dimensional logarithmic Gabor filter; obtaining a sea-land membership function by using an FCM method, and creating a global sea-land priori; constructing a cross-sea bridge marker field based on LSD, and creating a marker field priori of a non-water-permeable structure connected to an island; introducing a super-pixel layer to construct a water edge line detection model; inputting the single polarization SAR image into the water edge line detection model to obtain an output label of each pixel, and realizing island water edge line detection. The present application systematically solves the problems of coherent noise interference in the single polarization SAR image, difficulty in seed point initialization of the traditional random walk method, insufficient water edge line extraction precision caused by complex island contour, and invalidity of gray scale similarity in edge detection caused by large differences in ground object backscattering, and can effectively extract the island coast line of the single polarization SAR image.

The application provides a replay voice attack detection method for a voiceprint security authentication system. First, the voice signal is preprocessed, then a plurality of sub-band signals are obtained through linear equal-width Gabor filters, each sub-band signal is processed through FDEO to obtain an instantaneous amplitude and an instantaneous frequency, then the instantaneous amplitude and the instantaneous frequency are respectively taken as inputs of an SENet to obtain enhanced IACC and IFCC features, and the features are respectively processed through windowing and averaging and discrete cosine transformation to obtain respective low-dimensional feature vectors. Then, the extracted IACC and IFCC feature vectors are respectively used to train respective Gaussian mixture model classifiers to obtain respective classifier model parameters. In detection, the IACC and IFCC feature vectors of the voice to be detected are respectively input into the respective GMM classifiers and are scored for credibility, and finally score level fusion is performed to realize discrimination of true and false voices.

The invention provides a rock image classification method based on super-resolution and texture feature fusion. According to the method, the accuracy of mineral classification in the rock slice image is remarkably improved by combining a super-resolution model and a texture feature analysis technology, the enhanced image texture features are extracted by utilizing a Gabor filter, and classification is carried out by adopting a support vector machine. By comparing the performance of various SR models, the optimal model is selected to enhance the image resolution, so that the recognition effect of minerals and pores is improved. In addition, the system further optimizes the classification result through multi-color space feature fusion, finer mineral distribution data is provided for geological analysis, the mineral classification efficiency and precision are improved, and reliable technical support is provided for geological evaluation of oil and gas reservoirs.

The invention relates to the technical field of computer vision and image processing, and discloses a magnet surface defect detection method based on computer vision. Obtaining a grayscale image, and carrying out median denoising and linear normalization; a plurality of Gabor filtering kernels are generated in a multi-scale and multi-direction self-adaption mode, and convolution is carried out on the preprocessed image to obtain a response; extracting a phase and accumulating cosine and sine components to calculate an average phase; calculating phase consistency based on the phase deviation; performing adaptive threshold segmentation to generate candidate masks; carrying out 8-neighborhood connected domain analysis on the mask and filtering small-area noise; and calculating a bounding box, a centroid and an area of each effective connected domain and outputting a report. Through Gabor phase consistency measurement and self-adaptive segmentation, the problems of process fragments, parameter mismatch, missing and false detection and the like are solved, high robustness and accuracy are achieved, and the method is suitable for industrial automatic quality inspection.

The invention relates to the field of image classification, in particular to a hyperspectral image classification method based on a decoupling Gabor network. Two one-dimensional Gabor kernels along an orthogonal space axis are adopted, each one-dimensional Gabor kernel comprises a decoupling Gabor filter in the x direction and a decoupling Gabor filter in the y direction, a decoupling Gabor convolution module is constructed based on the decoupling Gabor filters, and specific redundant parameters of channels are greatly reduced while the effective characterization capability is reserved by sharing Gabor parameters and a parameter scattering scheme between the channels, so that the performance of the system is improved. And the overall complexity of the network is obviously reduced. And through a decoupling Gabor convolution module, a global average pooling layer and a multiple perceptron layer are combined to obtain a decoupling Gabor network. After input data is preprocessed, a network is trained, and finally, a hyperspectral image is put into the network to realize a classification effect. According to the method, efficient deployment is realized in a resource-limited hardware environment, the calculation cost and model parameters are remarkably reduced when high-dimensional data are processed, and the training and reasoning efficiency of the network is improved.

The invention discloses a metal curved surface grating three-dimensional contour macro-micro cross-scale characterization method and system. The method comprises the following steps: reconstructing global three-dimensional morphology data of a measured metal curved surface diffraction grating based on original data of sub-aperture splicing measurement of the measured metal curved surface diffraction grating; performing frequency domain filtering separation to obtain microscopic diffraction structure surface data and macroscopic substrate surface shape data; performing iterative nearest point surface shape registration on the macroscopic substrate surface shape data to calculate a grating substrate surface shape error; carrying out two-dimensional Gabor filtering on the surface data of the microscopic diffraction structure to calculate a mean value and a standard deviation of a grating period, and carrying out local plane least square fitting to calculate spatial distribution of the grating period and a blaze angle; and outputting the processing results of the macroscopic substrate surface shape data and the microscopic diffraction structure surface data as a three-dimensional contour cross-scale representation result. The invention aims to simultaneously realize high-precision characterization of the macroscopic substrate and the microscopic diffraction structure of the curved-surface grating aiming at the microstructure of the metal curved-surface diffraction grating.

The invention provides a method for predicting spinal refracture of an OVCF patient based on an X-ray image. According to the method, a multi-scale space-time convolutional neural network is adopted to analyze a standardized thoracolumbar spine X-ray image sequence, a vertebral body morphology analysis module, a microstructure texture feature extraction module and a time sequence dynamic modeling module are integrated, and the vertebral body morphology analysis module is used for extracting vertebral body geometric morphology features including the height ratio relation of the front edge, the center and the rear edge; the microstructure texture feature extraction module adopts a multi-direction Gabor filter combination and a local binary mode operator to extract the directivity, density and connectivity features of the bone trabecula; the time sequence dynamic modeling module adopts a hybrid architecture of combining a three-dimensional convolutional neural network with a long-short-term memory network; the system further comprises inter-vertebral-body relevance modeling. According to the method, a multi-task loss function item is adopted for joint optimization. According to the method, the prediction accuracy reaches 92.6%, patients can be divided into three risk levels, and effective decision support is provided for clinical precise prevention and treatment.

The invention provides a printing pattern retrieval method and device based on multi-feature fusion matching, equipment and a medium, and the method comprises the steps: processing an input printing image through an LBP operator, and obtaining the first texture information of the printing image; acquiring second texture information of different directions and frequencies of the printed image through a Gabor filter; fusing the first texture information and the second texture information to obtain fused texture information; the method comprises the following steps: setting an IP-Adter extractor to be in a stroke transfer mode, and extracting a corresponding style feature from a printed image; fusing the texture information and the style features to form a final feature vector; and carrying out similarity calculation on the final feature vector and the feature vector of each picture, and displaying the picture with the highest similarity, so that the required pattern is accurately retrieved, and the error detection condition is avoided.

The application discloses a bicycle tooth disc surface spraying quality detection method, the application detects the target area of the image quickly and accurately through the YOLOv3 model, and then extracts the texture feature information of the bicycle tooth disc spraying surface by using a Gabor filter.The method can effectively distinguish the surface features of different particles and capture subtle texture changes.Finally, the extracted texture features are input into a support vector machine (SVM) classification model for training and classification, so that the spraying quality is accurately evaluated.Compared with the traditional manual detection method, the application significantly improves the detection accuracy and consistency, avoids human errors, is especially suitable for complex situations where the particle size and surface roughness are difficult to distinguish, and provides more objective and accurate spraying quality judgment.

The invention discloses a high-precision intelligent super-resolution wavefront restoration method, and the method comprises the steps: designing a neural network model based on an optical imaging mechanism and a signal processing method, and building a nonlinear relation between a Shack-Hartmann wavefront sensor light spot array image and a Zernike coefficient through model training; a Gabor filter of which the bandwidth and the direction can be trained is used for replacing a convolution kernel of a traditional convolutional neural network, an SH-GaborNet model applied to a Shack-Hartmann wavefront sensor is constructed, high-frequency and low-frequency characteristics in dispersed light spots are efficiently separated and extracted, and high-resolution and high-precision wavefront restoration is achieved. According to the method, the limitation of spatial resolution is broken through, and the mode confusion error caused by inaccurate slope measurement is effectively inhibited; according to the method, higher precision is shown in a super-resolution wavefront reconstruction task, and an efficient and reliable technical solution is provided for the fields of wavefront detection with high precision and high resolution and the like.

The invention discloses a garment fabric sewing defect visual detection method, and relates to the technical field of visual detection, and the method comprises the steps: collecting an image data set, carrying out the preprocessing of the image data set to obtain a texture direction consistency feature set, calculating a texture direction consistency value and a main direction distribution set based on a multi-direction Gabor filter, and carrying out the visual detection of the garment fabric sewing defect. And judging the skewing or overlapping phenomenon of the sewing lines through the deviation ratio set, generating an abnormal mark set, calculating a sewing quality index in combination with the texture direction consistency value and the deviation ratio set, and finally outputting an alarm set and an adjustment instruction to a control system. According to the method, dynamic monitoring and quantitative evaluation of texture direction consistency in a sewing state can be realized, the detection precision and timeliness are remarkably improved, the problem of sewing state evaluation under a complex texture condition is solved, the technical transformation from passive monitoring to active protection is realized, and the working efficiency is improved. And a high-adaptability and high-practicability technical support is provided for efficient production in the textile and clothing industry.

The invention relates to the technical field of space surveying and mapping, and discloses a surveying and mapping platform and method based on big data, and the method comprises the steps: generating an initial pixel feature for a pixel of each SAR image; inputting the initial pixel feature into a first neural network, and outputting a parameter vector; generating Gabor cores through the parameter vectors to perform Gabor filtering processing on the SAR image, and obtaining filtered images with the same number as the Gabor cores; intercepting a filtering image of the target area on the auxiliary image and the main image to generate an area filtering image, inputting the generated area filtering image into a convolutional neural network, and outputting a deformation value of the target area by the convolutional neural network; a first neural network is trained to enable the first neural network to have the performance of adaptively generating a Gabor kernel for registering a current set of SAR images, then spatial differences of different SAR images are corrected through Gabor filtering, and assistance of high-precision DEM data does not need to be relied on.

The application discloses a kind of finger vein identification method, device, computer readable storage medium and equipment, belong to the field of biometric identification.It includes: obtaining finger vein image;Calculate the gradient information of finger vein image, according to gradient information, find the finger boundary using greedy strategy;Detect finger joint on finger vein image, and extract the effective area of finger vein according to the finger joint and finger boundary;Through Gabor filter, the effective area is extracted, and the finger vein feature is obtained;Set sliding window on the finger vein image, and the finger vein feature of the finger vein image in sliding window is compared with the corresponding position of the finger vein feature template Similarity, find the finger vein image in the highest similarity sliding window as the maximum matching area;The similarity of maximum matching area and the corresponding position of finger vein feature template is used as matching score.The application can realize stable and efficient finger vein identification for different devices and different collection environments.

The application belongs to the technical field of image data processing, and particularly relates to a large shaft forging blank center forging and pressing method and system based on machine vision, which comprises the following steps: pre-processing a high-temperature steel ingot image; adopting a logarithm-Gabor filter set to convolve the image, and obtaining a complex response and local direction energy; for each pixel point, obtaining its structural saliency based on the sum of the real part, the imaginary part and the local direction energy of the complex response; obtaining its direction certainty based on the local direction energy and the direction vector; obtaining the final voting weight by combining the structural saliency and the direction certainty; and based on the weight, accumulating the votes and calculating the steel ingot geometric axis, so that the anvil of the forging and pressing machine is aligned with the steel ingot geometric axis. The application effectively suppresses strong interference such as water vapor and oxide skin by combining the authenticity and directionality of the pixel points for weighted voting, and realizes stable and accurate positioning of the steel ingot axis under complex working conditions.

The invention belongs to the field of self-adaptive coal dressing, and particularly discloses a coal quality adaptive coal dressing method based on automatic energy amplitude adjustment, and the method comprises the steps of coal quality detection, coal quality evaluation, self-adaptive energy amplitude adjustment and sorting effect monitoring. According to the method, a multi-dimensional coal quality detection model is constructed, an RGB image is converted into a plurality of color spaces to extract comprehensive color features, after texture features are extracted from multiple frequency domains in combination with discrete cosine transform, discrete Fourier transform, discrete wavelet transform and Gabor filtering, quantitative analysis is performed by using an artificial neural network, and the coal quality detection accuracy is improved. High-precision detection of fixed carbon, ash content, volatile components and moisture is realized; through an adaptive control method based on meta-learning, a nonlinear regression network is established as an energy amplitude agent model, model parameters are optimized by using a gradient descent method to minimize an adjustment error, finally, an energy amplitude adaptive to the current coal quality is dynamically output according to a coal quality evaluation result, and full-automatic dynamic adjustment of the energy amplitude is realized.

The invention relates to the technical field of coast waterline detection, in particular to an island waterline detection method based on phase consistency random walk, which comprises the following steps of: converting a single-polarized SAR (Synthetic Aperture Radar) image into a phase consistency graph by using a two-dimensional logarithm Gabor filter; an FCM method is adopted to obtain a sea-land membership function, and sea-land global priori is created; constructing a cross-sea bridge marking field based on LSD, and creating a marking field prior of a non-permeable structure connected with an island; introducing a superpixel layer to construct a waterline detection model; and inputting the single-polarized SAR image into the waterline detection model, obtaining each pixel output label, and realizing island waterline detection. The method systematically solves the problems of coherent noise interference in the single-polarized SAR image, difficulty in seed point initialization in a traditional random walk method, insufficient waterline extraction precision caused by a complex island contour, and gray similarity failure in edge detection caused by large ground feature backscattering difference, and can effectively extract the island shoreline of the single-polarized SAR image.

The application relates to a bridge structure modal parameter identification method, which is based on a bridge structure micro-vibration video collected by a camera, fuses a variational mode decomposition (VMD), a phase-based video motion magnification (PBVM), a Gabor filter-based phase extraction method (GBP), an improved clustering algorithm (ICA) and an integrated covariance-driven stochastic subspace identification (SSI-Cov) and frequency domain decomposition (FDD) algorithm, and constructs a complete non-contact structure modal parameter identification system. In the absence of prior knowledge of the target frequency bandwidth, environmental noise interference and without obvious characteristic targets, the effective decoupling of multiple vibration modes of the bridge structure, the frequency domain feature extraction of small-amplitude vibration and the intelligent and automatic modal parameter extraction are realized, and the application belongs to the technical field of bridge structure health monitoring.

This application provides an anti-interference texture enhancement localization method, system, and medium suitable for low-texture visual scenes. The method includes: acquiring an original image; preprocessing the original image; extracting the texture response of the preprocessed image in different directions and frequency dimensions based on multi-scale Gabor filtering to obtain multiple texture response maps; dividing the preprocessed image into multiple image partitions based on a partitioned LED array and calculating the local gradient contrast of each image partition; independently adjusting the light source brightness of different image partitions to obtain an illumination-optimized image; performing feature fusion processing on the multiple texture response maps and the illumination-optimized image to extract feature points for visual localization; improving the consistency of images acquired at different time periods by normalizing the frequency response of the original image, histogram matching, and adaptive brightness restoration; and improving the local contrast by using a partitioned LED array with spatial illumination modulation to improve visual localization accuracy.

The invention relates to the technical field of vibration monitoring in mechanical dynamic control, in particular to a visual vibration measurement method based on phase zero crossing point and mark center detection, which comprises the following steps of: firstly, manually initializing an ROI (Region of Interest), then extracting cross curve parameters, and then, extracting a cross mark parameter based on a cross mark parameter; two pairs of complex Gabor filters are adopted to extract phase information of the cut cross image in two directions, cross mark center detection based on a phase zero crossing point is carried out, an intersection point of two zero crossing lines is obtained, namely the center point of a cross mark, and accurate and robust extraction of the center position of the cross mark is achieved; and the ROI position of the next frame is directly adjusted based on the center coordinate detected by the current frame, dynamic ROI updating is performed, and high-precision vibration monitoring of the target under the conditions of large-range translation, rotation and perspective distortion is realized.

The invention discloses a micro welding spot defect identification method based on computer vision, which comprises the following steps: S1, acquiring micro welding spot image data and preprocessing to generate a standardized image data set; s2, constructing a Gabor filter parameter space, and setting an initial value and a value range; s3, parameters are optimized based on a simulated annealing algorithm, Gabor features are extracted, and fitness is evaluated; s4, performing multi-scale and multi-direction filtering by using the optimal Gabor parameter to generate a high-dimensional texture feature vector; and S5, performing defect identification based on the high-dimensional texture feature vector, outputting defect types and positions, and performing feedback adjustment. According to the method, the simulated annealing algorithm and the Gabor filter are fused, so that efficient feature extraction and intelligent defect identification of the micro welding spot image are realized, and the detection accuracy and the automation level are improved.