422 results about "Gray level" patented technology

The invention relates to the technical field of size detection, in particular to a component size detection system based on machine vision, which comprises an interference area identification module, an edge repair module, a structure clustering module, a distortion analysis module and an attitude restoration module. According to the method, the gray level change trend of each pixel and the adjacent points in the image is obtained, the high-reflection suspected interference area is identified, the edge response cancellation processing is executed, and linear interpolation compensation is carried out by combining the effective edge response values of the adjacent areas; after a continuous contour trend is constructed, clustering and sub-grouping division are carried out according to an included angle difference trend between contour geometric vectors, a three-axis direction response linear array variation amplitude is extracted to construct a distortion curvature mapping table, existence of non-equal-interval deformation is judged, reverse angle compensation is applied, and a three-axis intersection point convergence process is completed through a nonlinear optimization algorithm; and according to the three-dimensional coordinates, calculating a geometric dimension and outputting a detection result, thereby realizing self-adaptive detection of different postures and surface interference component dimensions.

The invention relates to the technical field of image analysis, in particular to a vegetation fine classification and recognition method and system based on an unmanned aerial vehicle, and the method comprises the following steps: obtaining a multispectral image through the unmanned aerial vehicle, extracting red edge reflectivity, NDVI and gray-level co-occurrence contrast, generating a feature vector in a standardized manner, calculating neighborhood offset to obtain a dynamic weight, and combining the dynamic weight into a weighted vector; high discrete features are screened as effective channels, multi-scale clustering is carried out, center and region growth extension recognition is optimized, and a vegetation classification atlas is generated. According to the method, a neighborhood pixel feature offset dynamic weight mechanism is introduced, multi-spectral feature dimension contribution degree is adjusted in a self-matching mode, effective channels are screened based on full-image dispersion, redundant interference is eliminated, image pyramid multi-scale clustering and consistency constraint are fused, the complex vegetation boundary recognition capability is improved, dynamic weight and multi-scale optimization are coordinated, and the method is high in robustness and high in robustness. Sample dependence is reduced, and accurate distinguishing of spectrum similar vegetation is achieved.

The invention relates to the technical field of image analysis, in particular to a metal composite material surface defect intelligent detection method based on machine vision, which comprises the following steps: acquiring an original image and converting the original image into an observation matrix; decomposing the matrix into a low-rank matrix and a sparse matrix through a robust principal component analysis algorithm; utilizing a nuclear norm constraint low-rank matrix to establish a background statistical model, and utilizing a norm constraint sparse matrix; positioning a defect area in the sparse matrix and extracting a gray level co-occurrence matrix feature vector; performing connected domain analysis on the sparse matrix, calculating geometrical morphology parameters and evaluating a stress concentration coefficient; and establishing a self-adaptive decision model to carry out fusion judgment so as to judge the physical damage. According to the method, through low-rank sparse matrix decoupling and multi-dimensional statistical geometric feature fusion analysis, accurate stripping of weak defects and quantitative evaluation of physical damage attributes under a complex background are realized.

The invention relates to the field of computational imaging, in particular to a computational imaging method of a crossed strip-shaped anode single-photon detector, which comprises the following steps of: screening effective data through full-view-field illumination, and calculating an average output signal amplitude of each anode strip and a total average output signal amplitude of all anode strips to obtain a correction factor; the method comprises the following steps of: correcting an original output signal of each anode strip by adopting a correction factor during target imaging, performing effective signal data screening on the corrected original output signal, substituting effective signal data into a Gaussian algorithm to calculate an incidence position coordinate, mapping the incidence position coordinate to a mapping matrix, and stretching the incidence position coordinate to a 256 gray level for display. According to the calculation imaging method provided by the invention, the fixed mode noise can be suppressed, the imaging resolution is improved, and the image distortion is reduced.

The invention discloses a motor insertion piece positioning detection method using optical recognition, and particularly relates to the technical field of motor manufacturing and detection. Collecting a high-resolution image containing the insert structure; carrying out image gray level enhancement and reflection interference suppression; extracting insert edge gradient information, and constructing an insert boundary point set; fitting the profile of the insert according to the boundary point set, and correcting the profile through a shape prior model; extracting barycentric coordinates and center positioning features of the inserts, calculating offsets delta X and delta Y and an angular deviation theta, and constructing a position offset vector V; the V is compared with a set tolerance model T, and whether the insertion piece is qualified or not is judged; an error distribution thermodynamic diagram is generated based on the offset data of the multiple punching sheets, and trend evaluation of the mold state and the material stability is achieved; according to the method, high-precision, full-automatic and visual detection of the sheet inserting position can be realized, and the sheet punching quality control level is improved.

The invention relates to a grayscale request routing method and device based on a message queue, equipment and a medium. In the application, after a message production service receives a request message, if a message agency judges that the request message carries a gray level identifier, an original theme of the request message is modified into a shadow theme to generate a target message, and the target message is sent to a gray level message queue, so that a gray level service in a message consumption service is facilitated; and consuming the corresponding target message from the grayscale message queue through the shadow theme. Visibly, according to the method, the device and the system, the grayscale request carrying the grayscale identifier is identified through the message agent, the original theme of the grayscale request is modified into the shadow theme, and then the shadow theme is stored in the grayscale message queue, so that grayscale request routing based on the message queue can be realized, the grayscale request can be isolated from other online requests in the mode, and the grayscale traffic is prevented from influencing the online user traffic; the message agent is independent of the business program of the message production service, so that the scheme does not need to modify the business code, and the stability of the business code is ensured.

The invention discloses a brightness and chrominance consistency correction method and system for an LED spliced screen, and the method comprises the steps: firstly recognizing the batch features of a module, dividing the batch features into correction partitions, and respectively collecting the original brightness and chrominance data of a high-gray linear region and a low-gray nonlinear region; and then, calculating the geometric intersection of the original color gamuts of all the partitions to determine the maximum common target color gamut, and eliminating the metamerism phenomenon caused by multi-batch mixed use from the physical level. Then, constructing a non-linear dual-order correction model, and generating a correction coefficient set containing a main linear conversion matrix and a low-gray-scale non-linear offset item; when the control system runs, the linear matrix is dynamically called according to the input gray level, and bias compensation is superposed. According to the invention, the problems of cross-batch visual angle chromatic aberration and color temperature drift and pocking marks under low gray level of the spliced screen are effectively solved, and the high-fidelity display consistency of the full gray level is realized.

The invention discloses a GLCM texture feature extraction method and system based on dynamic multi-scale weighting, and relates to the technical field of image feature extraction, and the method comprises the steps: carrying out the image preprocessing and texture scale perception of an input image; carrying out the precise adaptation of the pixel distance and the texture scale of the processed image, dynamically adjusting the gray level of a gray level co-occurrence matrix GLCM according to the image texture complexity, and constructing a dynamic multi-scale GLCM; weight self-adaptive distribution is carried out on each texture scale and direction of the dynamic multi-scale GLCM, and original feature extraction is carried out in combination with distributed weights; and carrying out weighted fusion on the extracted original features to obtain fusion features, and carrying out anti-noise optimization to obtain standardized texture features. A scale parameter dynamic adjustment and feature weighting mechanism is adopted, the problems that a traditional GLCM is poor in adaptability to a complex texture scene and insufficient in feature discrimination degree are solved, and the characterization capacity of texture features to a measured sample and the feature discrimination degree and robustness in the complex texture scene are improved.

The invention relates to the technical field of image recognition, in particular to a urinary system CT image tumor benign and malignant identification method based on deep learning, which comprises the following steps: acquiring a urinary system CT image, constructing a map band sequence and extracting gray level distribution, identifying a heterogeneous edge and a texture mutation region, aggregating perturbation map blocks to form an abnormal structure, and identifying the benign and malignant tumors. And fusing multiple types of image layers to complete label integration, and generating a feature recognition image layer. According to the method, the extension recognition capability of the tumor edge external expansion region is enhanced by combining a graph band gray scale aggregation and sequence construction mode, the judgment precision of local heterogeneous change is improved by fusing gray scale kurtosis and migration analysis, and the texture disturbance trend is extracted based on direction vector included angle change. The block gray level fluctuation and gradient relationship supports abnormal structure aggregation identification, spatial coincidence and boundary difference combined screening realizes multi-feature region unified coverage, abnormal region expression definition and structure positioning accuracy are enhanced, and image layer consistency and identification stability are improved.

The invention belongs to the technical field of fault diagnosis, and discloses a thermal defect identification method and system for a high-voltage switchgear, and computer equipment, and the method comprises the steps: firstly segmenting an infrared image through employing a transfer learning optimized Mask R-CNN model, reducing the dependence of annotated data through sharing pre-training parameters, and achieving the region extraction of pixel-level equipment; secondly, multi-dimensional temperature information is extracted in combination with a gray histogram and a gray co-occurrence matrix, and key features are screened through PCA to enhance noise immunity; and finally, the LSSVM is adopted for classification, so that the training efficiency is remarkably improved. According to the method, the equipment area is automatically segmented through deep learning, temperature distribution is quantified in combination with multi-dimensional features, man-made misjudgment is reduced, pre-training model parameter sharing is utilized, new tasks are rapidly adapted in a small sample scene, the generalization ability and efficiency are improved, feature dimensions are compressed through principal component analysis, the real-time monitoring requirement is met, and the monitoring efficiency is improved.

The invention relates to the technical field of computer vision, in particular to a text-guided creativity and time consistency-oriented video coloring method and system. By introducing a time deformable attention block, according to the position and shape change dynamic state of an object in a video frame, dynamic features of an instance are captured in the time dimension, so that feature representation consistency is kept, and color flickering and shifting in the coloring process are effectively prevented; semantic representation of a text noun concept is adjusted through a cross-modal pre-fusion module, mask cross attention in the module is used for enhancing understanding of the model on the noun concept and instance perception, and color distribution accuracy is improved; the global structure of the colored video is maintained through gray level video information introduced in the video decoding process; a cross-fragment fusion mechanism is introduced during reasoning, so that the long-term consistency of video coloring is maintained; through text guidance, the user can color the video according to own demands and creativity.

A cockpit display system includes a cockpit, a first display apparatus, a first light sensor, a second light sensor and a brightness distribution calculation module. The first light sensor is suitable for detecting a first ambient light brightness. The second light sensor is suitable for detecting a second ambient light brightness. The brightness distribution calculation module is suitable for respectively calculating a first brightness and a second brightness of the first display area and the second display area of the first display apparatus under a same display gray level according to the first ambient light brightness and the second ambient light brightness. The first brightness and the second brightness are different from each other.

The invention relates to the technical field of machine vision detection, in particular to an automobile part surface defect online detection method and system. The method comprises the following steps: acquiring a grayscale image, calculating the size of a structural element for each pixel based on a local Gaussian Laplacian operator response variance, filtering to obtain a substrate image according to the size of the structural element, and differentiating to obtain a texture image. Determining a Gabor scale and a gray-level co-occurrence matrix statistical direction by using the size, and extracting a cooperative direction gray-level co-occurrence matrix feature; and a weight is set based on the size and is subjected to weighted fusion with a multi-scale rotation invariant local binary pattern feature to generate a texture saliency map, and texture defects are judged. On the substrate image, taking the gray value as the height, and determining a neighborhood calculation curvature feature based on the size to detect the substrate defect. According to the scheme, the image scale can be adaptively analyzed, the background texture is effectively inhibited, and therefore different types of tiny defects such as scratches and pits can be reliably detected.

The invention provides an ultrasonic image focus automatic identification method based on deep learning, and the method comprises the steps: obtaining an ultrasonic probe pressure and an ultrasonic image of a body surface contact region of a patient, carrying out the tissue hierarchy analysis of the ultrasonic image, and recognizing the tissue density and the three-dimensional space coordinate information of the depth position of a target organ; driving sound wave signals of the ultrasonic probe through the target probe frequency, monitoring the contact pressure change of the probe and the body surface in real time, and judging whether the current contact state meets the pressure requirement of deep imaging or not according to the contact pressure change; identifying a deep focus boundary of the enhanced ultrasonic image, and performing multi-scale feature extraction and texture analysis on the ultrasonic image to obtain contour coordinates, tissue gray level distribution and internal uniformity of a focus area; meanwhile, the image noise level and the tissue contrast difference are evaluated to obtain the definition, the contrast ratio and the boundary sharpening degree of the current image.

The invention relates to the technical field of image processing, in particular to an online calibration method and system for hydrological flow data, and solves the technical problem of low flow calculation precision caused by rough window segmentation and illumination interference in the prior art. The method comprises the following steps: collecting a water surface video in a river water flow process, and setting a velocity measurement line area in the water surface video along a water flow direction; according to the gray level distribution of the velocity measurement line area, carrying out illumination state judgment, and according to a judgment result, carrying out image restoration on an illumination abnormal area in the water surface video to obtain a restored water surface video; constructing a space-time image for the repaired water surface video, performing corner detection, clustering corners to divide a plurality of windows, and constructing a window weight for each window; and weighting the texture angle of each window according to the window weight of each window to obtain a weighted texture angle, and calculating hydrological flow data according to the weighted texture angle.

The invention relates to the technical field of image processing, in particular to a flower disease and insect pest monitoring method based on machine vision, and the method comprises the steps: obtaining a candidate pseudo boundary set in a gray level image of a to-be-monitored flower leaf, and obtaining a gray level distribution feature of a pixel point in a local analysis region of each candidate pseudo boundary; obtaining a vein feature index of each candidate pseudo-boundary according to the linear structure feature of each candidate pseudo-boundary and the linear structure feature of each candidate pseudo-boundary; according to the vein feature index and the direction feature of each candidate pseudo boundary, screening in the candidate pseudo boundary set to obtain at least one pseudo boundary; in the process of carrying out region segmentation on the gray level image by using the watershed algorithm, the initial distance of each pixel point after distance transformation is obtained, adaptive distance compensation is carried out on the initial distance of each pixel point according to the position relation between each pixel point and each pseudo boundary, the compensation distance is obtained, complete segmentation of a scab region is realized, and the scab image segmentation efficiency is improved. And a reliable quantitative basis is provided for flower disease and pest monitoring and prevention decision making.

The invention discloses a Java enterprise-level rapid development method based on visual modeling. The method comprises the steps that modeling behaviors are collected to generate a cognitive map, a view is constructed, and operation behaviors are collected to generate cognitive feature data; generating a semantic constraint model, and fusing the cognitive features and the node features to construct the semantic constraint model; semantic consistency detection is executed, design and implementation differences are automatically compared, and bidirectional correction is triggered; generating a project engineering structure, and generating a structured code and an interface document according to the correction model; zero-shutdown migration evolution is executed, a migration relation graph is constructed, and seamless switching is carried out in a gray level mode; and performing semantic self-correction updating in the operation period, monitoring semantic offset and backfilling design content for closed-loop synchronization. According to the method, intelligent modeling and uninterrupted evolution of Java enterprise-level applications are realized through cognitive map driving and a multi-mode migration mechanism.

The invention relates to the technical field of intelligent detection and automatic control, in particular to a coal moisture content intelligent sensing and dust suppression control method based on hyperspectrum, which comprises the following steps: acquiring an image and sorting a gray interval, removing a high-anti-interference area, segmenting numbered blocks, counting a gray direction reversal identification interference area, extracting peak positions of each wave band, and calculating difference value groups. And aggregating the continuous image blocks to extract gray boundaries, matching the boundaries and spraying serial numbers to adjust and control, and outputting the moisture content and a dust suppression image. According to the method, a region segmentation method driven by reflection density is introduced, gray level warping and interference contour elimination are combined, a block set is constructed to enhance the processing pertinence, through multi-band peak position difference value and symbol sequence continuity judgment, peak position change grouping is established to identify the spectrum trend, boundary aggregation is completed in the gray level mean value difference value direction, and finally, the spectrum trend is identified. And a control chart is generated by combining serial number position information, and linkage of an image structure, spectral distribution and control feedback is realized.

The invention relates to the technical field of machine vision and industrial quality control, and discloses a cigarette packet surface printing defect detection method and system based on vision intelligence. The method comprises the following steps: obtaining an initial image of the surface of a cigarette packet, and adaptively adjusting optical parameters to obtain a clear image; pixel-level features are extracted, gray level conversion is carried out, and a potential defect area boundary is positioned; identifying edge distortion features and classifying font deformation types; separating a target pixel block, matching the target pixel block with a standard template, and quantifying color deviation; carrying out ambiguity evaluation on the deviation overrun region, and determining a fuzzy position; extracting the sub-image blocks and performing cross validation in combination with the classification labels to generate a comprehensive defect description vector; and generating a real-time feedback signal in the high-speed production line according to the vector, dynamically adjusting imaging parameters, and realizing continuous and accurate defect identification. The method can effectively improve the detection precision and adaptability of the cigarette packet surface printing defects, and is suitable for real-time quality control in a complex industrial environment.

A display device, which fixes the magnitude of a driving current flowing through light-emitting devices in a low luminance region and controls an application period of the driving current, includes: light-emitting devices; a processor for controlling the light-emitting devices on the basis of a gray level required for the light-emitting devices in each unit frame; and a timing controller which divides the unit frame into a plurality of sub-frames and generates a scan signal corresponding to each sub-frame. The processor: when the gray level is greater than or equal to a preset value, determines a luminance value of the light-emitting devices on the basis of the gray level and turns on the light-emitting devices with the determined luminance value in all of the plurality of sub-frames; when the gray level is less than the preset value, determines a light-emitting period of the light-emitting devices on the basis of the gray level and divides, on the basis of the light-emitting period, the plurality of sub-frames into a first and a second sub-frame group; turns on the light-emitting devices with a preset luminance value in all sub-frames belonging to the first sub-frame group; and turns off the light-emitting devices in sub-frames belonging to the second sub-frame group or turns on the light-emitting devices with a luminance value less than the preset luminance value.

This application provides a brightness compensation method for a display panel. The method includes: when detecting that the display panel displays a to-be-displayed image at a current display brightness value, determining a brightness interval in which the current display brightness value is located; determining gray level intervals corresponding to the brightness interval; and obtaining a gray level of the to-be-displayed image, and compensating each pixel unit in the display panel based on a brightness compensation value corresponding to a gray level interval in which the gray level is located.

The invention relates to the technical field of video generation, in particular to a video frame generation system and display equipment based on texture reconstruction and trajectory optimization, and the system comprises a motion breakpoint detection module, a trajectory interpolation frame generation module, a texture anomaly positioning module, a texture interpolation reconstruction module and a video frame synthesis module. According to the method, an edge feature map is generated through edge detection and non-maximum suppression, the boundary recognition precision is improved, noise is suppressed, candidate regions are divided through region growth, the local feature extraction efficiency is enhanced, global redundancy is reduced, three-frame optical flow analysis captures trajectory breakpoints, and motion vector correction path deviation is predicted in combination with a Lucas-Kanade method. Color consistency verification optimizes interpolation to reduce cross-frame abrupt change; 64 * 64 grids and a gray level co-occurrence matrix locate texture abnormity; five-frame similarity strengthens time sequence stability; Gabor filtering extracts a main direction and combines an adjacent frame reconstruction texture direction to ensure consistency; and trajectory smoothness and visual consistency are optimized through multi-frame cooperation and layered detection.

A molten pool image processing method in the laser powder bed melting process comprises the steps that the center of a nominal molten pool is positioned, a region of interest is extracted, an optimal gray threshold value is calculated through a self-adaptive threshold value algorithm, binary processing is conducted on a gray image, and a binary molten pool image is obtained; performing edge detection on the binary molten pool image to obtain an edge image, constructing an edge point polar coordinate sequence, and calculating a polar angle difference matrix and a polar radius difference matrix; calculating the number of contours in the edge image, and judging whether the edge image has splashing points or not; calculating a polar angle difference value of adjacent edge points, and judging whether the contour of the molten pool is closed; on the premise of ensuring that the contour of the molten pool is closed, based on the shape symmetry assumption of the molten pool, the distance between symmetric points of the molten pool is calculated, basic geometrical characteristics of the molten pool are extracted, and physical values of the basic geometrical characteristics of the molten pool are calculated. Based on the gray level and spatial distribution difference of the molten pool and the background noise, the challenge of tracking the morphology of the molten pool under any processing condition in the actual production process can be relieved.

The invention discloses a fabric texture fidelity rendering method and system based on a scanned image and a target color number, and solves the technical problems of poor color perception consistency and texture detail loss. The method comprises the steps of preprocessing a standardized input image through illumination compensation, performing double-space color mapping in an LAB color space, partitioning the image by adopting K-means clustering according to texture features of a gray-level co-occurrence matrix, establishing a color mapping function containing global mapping transformation and local adaptive adjustment, and adaptively determining a fusion coefficient according to regional variance; decomposing the image into a base layer and a detail layer by adopting guide filtering, and performing spectrum constraint on the detail layer to ensure that the spectrum amplitude is consistent with the original; extracting material parameters such as a normal map, a roughness map and a metallity map; according to the method, fabric texture details are completely reserved while color perception consistency is guaranteed, rendering reality is improved, batch processing of multiple groups of target color numbers is supported, and the method is suitable for application scenes such as costume design, e-commerce display and virtual fitting.

The invention relates to the technical field of image recognition, in particular to a bridge cable protection layer damage recognition method and system based on machine vision, and the method comprises the steps: extracting a region of interest based on the edge information in a cable protection layer image, obtaining a gray histogram, and carrying out the recognition of the damage of a bridge cable protection layer through the gray value level of a wave crest in the gray histogram; determining a distinguishing value of pixel points in the region of interest; calculating a pixel mutation index of the pixel point by using the gray level change condition of any pixel point in the region of interest in different directions in the local range and combining the distinguishing value of the pixel point; calculating the characteristic value of the pixel point by combining the pixel mutation index of the pixel point and the gray level characteristic in each direction in the local range corresponding to the pixel point; and carrying out damage area identification by utilizing the characteristic values of all the pixel points in the region of interest and combining a machine learning algorithm. According to the invention, the damage identification accuracy of the bridge cable protection layer is effectively improved.

The invention provides a real-time foreign matter detection method and system based on multispectral imaging, and belongs to the technical field of computer vision and image processing.The method comprises the steps that multiband spectral images in the 400-1000 nm range of a target area are collected, noise sensitive bands are screened through a gray-level co-occurrence matrix, and targeted noise reduction is conducted; constructing a multispectral feature fusion network containing an attention mechanism to generate a fusion feature map; an improved YOLOv5 algorithm (including a feature pyramid attention module with an extrusion-excitation module) is adopted to detect foreign matters, and a result is output after post-processing. The system comprises a multispectral imaging module, a preprocessing module, a feature fusion module, a foreign matter detection module and a result output module. The foreign matter distinguishing capability is improved by utilizing the multispectral characteristic, real-time accurate detection is realized through targeted noise reduction, dynamic characteristic fusion and an optimization algorithm, and the method is suitable for the fields of industrial production, food processing and the like.

The invention discloses an image real-time rendering processing method in a digital twinborn scene, which relates to the technical field of real-time rendering and comprises the following steps: a multi-source data preprocessing step: noise reduction is carried out on an RGB image of the digital twinborn scene by adopting self-adaptive Gaussian filtering which dynamically adjusts the size of a filtering kernel according to the pixel gray level difference, and the noise of the RGB image is reduced; the depth image is processed through bilateral filtering with the coefficient of 0.8-1.2, and lightweight data processing is carried out, specifically, a unified world coordinate system is established, and three-dimensional model vertex data and image pixel coordinates are registered, and the method has the advantages that through differential noise reduction, unified coordinate registration and lightweight processing, data distortion and offset are reduced, and the precision of the image is improved. Fusion efficiency is improved, rendering delay is solved, and scene change and hardware difference are adapted by collecting hardware and performance parameters in real time, establishing an intelligent adjustment mechanism and matching scene switching preloading and transition frames.

The invention discloses a real-time obstacle sensing method for a sonar image, and relates to the technical field of sonar image processing, and the method comprises the following steps: collecting a time sequence sonar image flow of a target water area, carrying out the inter-frame alignment processing of the time sequence sonar image flow, and screening out an effective sonar image with a signal-to-noise ratio meeting a preset threshold value; a suspected obstacle area in the effective sonar image is extracted, and morphological features, gray level distribution features and texture features of the suspected obstacle area are obtained to form a suspected feature database; and performing similarity matching on standard features in the suspected feature database and the obstacle feature database to obtain a feature matching degree, so as to intuitively present the position, the type and the movement track of the obstacle by constructing an obstacle real-time distribution map under a three-dimensional coordinate system, and integrate a safety distance early warning mechanism, thereby improving the safety of the obstacle. And timely and comprehensive obstacle information support can be provided for underwater operation or navigation, and the operation safety and efficiency are ensured.

The invention discloses a building template cleaning monitoring method and system based on image recognition, and the method comprises the steps: building a smudge segmentation neural network model through a U-Net network, employing a ResNet-50 network to extract deep features through an encoder, introducing an attention gating mechanism into a decoder to enhance the capturing capability of features of tiny dirt, and carrying out the recognition of the tiny dirt. Inputting the initial building template image into a smudginess segmentation neural network model to obtain a semantic segmentation mask image; using a GLCM gray level co-occurrence matrix to identify a residual dirt area of the semantic segmentation mask image, extracting morphological, color and texture features, and calculating a dirt comprehensive severity index; and comparing the comprehensive smudginess severity index with a preset threshold value, evaluating the cleaning state of the building template according to a comparison result, and outputting an evaluation result. Quantitative evaluation of the smudginess degree is achieved, manual subjective misjudgment is avoided, and the monitoring precision and efficiency are improved.

The invention relates to the technical field of visual detection, in particular to a PCB defect visual detection method and system based on feature extraction, and the method comprises the steps: carrying out the evaluation of each cluster based on the pixel difference value and local density of an abnormal point on a to-be-detected PCB image, and the distance from the abnormal point to a clustering center, so as to determine an enhanced pixel value; and respectively comparing the enhanced pixel values of all the abnormal points in the row and the column where the abnormal points are located with corresponding pixel values in the qualified PCB image, and determining a pixel fluctuation amplitude by combining the abnormality degree so as to determine a pixel enhancement factor, so that the to-be-detected PCB image is enhanced for defect detection. According to the method, the contrast ratio of the slight defects in the PCB image is adaptively enhanced, the problem that a traditional gray level co-occurrence matrix method is not sensitive to detection of the tiny defects is solved, and the detection precision of the slight defects of the PCB is improved.