84 results about "Feature saliency" patented technology

The invention relates to the technical field of power equipment monitoring, and discloses a hydropower station high-altitude equipment fault intelligent identification system based on multi-sensor fusion, which collects multi-modal data in real time and evaluates data quality by deploying multi-source sensors at key parts of high-altitude equipment. Extracting multi-scale features of each modal, performing normalization processing, calculating a fusion weight based on feature saliency and data credibility, and performing weighted fusion and dimension reduction on the multi-modal features; based on the fusion feature vector, intelligent matching analysis of fault features and intelligent identification of fault types are carried out; a fault identification result is obtained; in addition, the system also comprises safety monitoring of overhead working personnel, and realizes closed-loop management from fault identification to safety maintenance. According to the invention, early weak faults can be accurately identified, and the safe operation level of equipment and the intelligent degree of operation safety management are improved.

The invention discloses a deep concealed ore deposit detection method based on multi-source geological data fusion, and the method comprises the following steps: 1, carrying out a dynamic multi-source data fusion algorithm: introducing a fuzzy logic algorithm, dynamically adjusting the data fusion weight according to the feature significance of different geological units, and developing a cross-scale feature pyramid network, hierarchical feature extraction and fusion are carried out on different detection depth data; 2, optimizing an intelligent model based on prior knowledge; step 3, constructing a real-time interactive three-dimensional modeling system; 4, establishing an efficient multi-means verification system; and step 5, intelligent integration of the exploration process. According to the method, multi-source data are fused through fuzzy logic dynamic weight adjustment, the mineralization mode and the reinforcement learning optimization model are combined, three-dimensional modeling and multi-means combined verification are updated in real time, and the effects of accurate fusion of the multi-source data, dynamic model adjustment and optimization, efficient and accurate modeling, short verification period and low cost are achieved; and the detection efficiency and scale of the deep concealed ore deposit are improved.

The invention discloses an image region analysis method based on entropy driving feature enhancement, and relates to the technical field of image analysis and feature enhancement. According to the method, the local channel information entropy is used as a core feature statistical magnitude, and adaptive weighting and strengthening of different importance region features are realized through explicit quantification of image feature information amount; weight distribution is dynamically adjusted according to the characteristic values, the response of a high-information dense area is remarkably enhanced, and meanwhile low-information and noise interference areas are effectively restrained. On the basis, a cross-layer attention mechanism based on entropy prior is designed, feature statistical information is embedded into a gating and weight generation process, and attention distribution with feature significance as guidance is achieved. According to the method, through an entropy-driven adaptive feature enhancement mechanism, the perception capability, the feature discrimination capability and the analysis precision of the model on the salient region of the image are effectively improved.

The invention belongs to the technical field of rehabilitation assessment, and provides a cerebral apoplexy rehabilitation assessment method and system based on multi-mode electroencephalogram and myoelectricity fusion. According to the method, 59-channel electroencephalogram signals and 14-channel electromyographic signals are preprocessed, mutual information between channels is calculated to construct a correlation matrix, spatial-temporal features are extracted in combination with a convolutional neural network (CNN) and a Transform self-attention mechanism, and rehabilitation level classification is achieved. According to the method, coherence between electroencephalogram / myoelectricity channels is quantified by adopting mutual information, and 59 * 59 and 14 * 14 dimensional feature matrixes are constructed; a CNN-Transform hybrid model is designed to optimize feature fusion, and the classification precision is improved; gradient weighted class activation mapping (Grad-CAM) is introduced to analyze feature saliency, and correlation between an electroencephalogram channel and a focus is revealed; and developing a visual evaluation system, and displaying the multi-modal data and the historical trend in real time. The system stores patient information and evaluation results through a MySQL database, and supports doctor-patient online interaction. Compared with traditional scale evaluation, the method has the advantages that the evaluation objectivity is improved by using multi-modal objective data, the model credibility is enhanced by combining interpretability analysis, and technical support is provided for formulating a personalized rehabilitation scheme.

The invention discloses an oil and gas pipeline magnetic flux leakage image defect identification method based on a deep attention mechanism, relates to the technical field of oil and gas pipeline detection, and is used for solving the problem of inaccurate identification of a magnetic flux leakage image of a pipeline elbow section. According to the method, the image frame sequence with the posture annotation is constructed through unified time reference and space coordinate mapping, and accurate alignment of the image and the pipeline position is achieved; a structural area marking graph and non-rigid normalization are introduced to compensate the distortion of the elbow section, and the image consistency is improved; constructing a structure perception embedded image on the distortion compensation image, fusing position, gradient and texture features to implement feature propagation and attention guidance, and generating a feature saliency map; a defect area is accurately extracted in combination with layered reconstruction and a two-stage judgment strategy; through continuous frame monitoring and recognition stability judgment, recognition parameters are adaptively updated, a closed loop from recognition to updating to verification is constructed, and the precision of oil and gas pipeline magnetic flux leakage image recognition under complex working conditions is enhanced.

The invention relates to the technical field of tunnel engineering monitoring and construction, and discloses a shield construction tunnel full deformation prediction method based on artificial intelligence. The method comprises the following steps: acquiring a geological sensing data stream, a construction operation data stream and a historical deformation case library; performing multi-modal data fusion on the geological sensing data flow and the construction operation data flow to generate a construction environment map with consistent time and space; searching a case sequence matched with the current construction environment map from a historical deformation case library, performing feature significance evaluation through the case sequence, and identifying a dominant feature group of full deformation prediction; constructing an algorithm preferential engine based on the dominant feature group; activating an algorithm preferential engine to perform parallel processing on the construction environment map, and generating an adaptability measurement set of each algorithm; integrating the adaptability measurement set and real-time construction limiting conditions, and selecting an optimal prediction algorithm by adopting a trade-off decision-making mechanism; and generating a full deformation prediction model, and integrating the model to a shield construction monitoring platform.

The invention relates to the technical field of virtual reality cultural relic construction, in particular to a cultural relic three-dimensional construction VR display system based on multi-modal data fusion. The specific implementation process comprises the steps of obtaining a laser point cloud and a two-dimensional image of the surface of a cultural relic, respectively generating a geometric point cloud and a visual sparse point cloud, and inputting the geometric point cloud and the visual sparse point cloud into a modal registration model based on geometry-luminosity combined constraint to complete pixel alignment and data fusion; performing texture mapping by using a Markov random field to generate a three-dimensional mapping model, generating a feature saliency map based on curvature calculation, and performing differential mesh simplification on flat and detail areas to output a polygonal contour; and generating a normal and a replacement map by using a ray casting algorithm, and importing the normal and the replacement map into a virtual reality engine for rendering display in combination with the polygonal contour. According to the method, a multi-mode joint optimization and differentiation grid simplification technology is utilized, dislocation and ghosting of texture mapping are effectively eliminated, and while VR rendering fluency is guaranteed, microscopic geometric details and physical material attributes of cultural relics are reserved.

The invention relates to the technical field of safety monitoring, in particular to a personnel safety intelligent monitoring system which comprises a millimeter wave monitoring unit and a real-time monitoring module of an image acquisition unit. The label setting module is used for calculating an abnormal characterization value and setting a potential abnormal label; the feature extraction module is used for estimating an image acquisition unit associated with the target and an associated time domain segment; the frame extraction analysis module is used for determining a video segment needing to be extracted; the feature analysis module is used for evaluating the feature prominence of the target in each time domain sub-segment; and the result analysis module is used for determining a time domain sub-segment needing to be analyzed based on an evaluation result of the feature analysis module for feature saliency, determining an action type of a target in the time domain sub-segment, and judging whether the target is abnormal or not according to a historical sample corresponding to the action type. According to the method, the potential anomaly analysis is carried out on the target, the target with characteristic highlight is selected, whether the anomaly exists or not is judged, and the anomaly monitoring efficiency and accuracy are improved.

The invention relates to the technical field of image processing, and discloses an automobile part defect detection method, which comprises the following steps of: acquiring two-dimensional image data and three-dimensional point cloud data of an automobile part, generating a depth mapping graph from the three-dimensional point cloud data, and splicing the depth mapping graph with a color channel of the two-dimensional image data to form a multi-channel input tensor; the multi-channel input tensor is fed into the feature extraction network to generate a feature saliency map, and the feature saliency map is utilized to guide a deformable convolution module to perform adaptive sampling and convolution on a feature map; performing multi-task prediction on the defect candidate area through an anchor-frame-free detection head; and constructing a topological association graph of the defect candidate regions, calculating a connection weight between nodes, pruning the graph by using a preset weight threshold, and merging the defect candidate regions into a final defect detection result. According to the method, the detection limitation of a single data source under complex illumination and diversified defect forms is overcome, and the comprehensive detection capability and detection stability of various defects are improved.

The invention discloses a cutterhead tunnel face instability prediction method based on an LSTM deep learning model, and belongs to the technical field of tunnel construction safety early warning. According to the method, a set of complete shield construction safety early warning system is constructed by fusing multi-source sensor data acquisition, feature significance analysis, a mixed database architecture and a time sequence prediction modeling technology. According to the system, a hybrid architecture combining structured and document type heterogeneous storage modes is adopted, a double-channel processing mechanism of equipment state data and time sequence characteristic data is established, and efficient integration of data is achieved through sliding window sampling and standardized preprocessing. A time sequence prediction algorithm based on an LSTM deep learning model is combined with a feature importance screening mechanism, so that the prediction precision and real-time performance of the instability state of the cutterhead tunnel face are remarkably improved. Experimental results show that the method provided by the invention can effectively capture dynamic characteristic changes in the shield construction process, and a reliable intelligent early warning solution is provided for safe construction of tunnel engineering.

The invention discloses a method and a system for quickly searching and positioning a target in a surveillance video, and the method comprises the steps: carrying out the large-class marking of the target in a key frame of the surveillance video, and generating first marking data; performing coarse classification marking on the targets of each large class according to the first marking data to generate second marking data; performing fine classification marking on the roughly classified target according to the second marked data to generate third marked data; generating a hierarchical association index; calculating feature saliency; selecting the key frame with the highest feature saliency as a retrieval display picture; generating a retrieval video clip; when the retrieval input is a character, generating a first matching result; when the retrieval input is a picture, generating a second matching result; and retrieving the corresponding retrieval display pictures and the retrieval video clips from the hierarchical association index, and outputting the retrieval display pictures and the retrieval video clips according to a sequence of similarity from high to low. The method and the device are used for improving the accuracy and the efficiency of searching and positioning the target in the monitoring video.

The invention relates to the technical field of computer vision, in particular to a fabric color fastness analysis method and system based on image processing. Comprising the following steps: acquiring a multispectral image of a to-be-detected fabric, and synchronously acquiring multi-factor data; processing the multispectral image to generate an enhanced image; performing image recognition on the gray sample card in the enhanced image to generate a calibration image; performing spectral data projection on the calibration image through a dimension reduction algorithm, and performing color analysis and quantification on color change to generate color difference features; extracting texture features by using a conditional generative adversarial network, carrying out image recognition and calculation on the calibration image through a decoupling algorithm, and generating a region credibility graph; and inputting the chromatic aberration features, the texture features, the multi-factor data and the regional credibility map into an image feature fusion model for mapping, performing analysis through gradient visualization, and outputting an abnormal feature saliency map. According to the method, a color fastness objective analysis closed loop is created, so that the accuracy and the universality of an analysis result are improved.

The invention discloses a historical block courtyard boundary intelligent extraction and semantic analysis method based on space-air-ground multi-dimensional sensing fusion, and the method comprises the steps: data fusion, the analysis of existing data, and the fusion of multivariate heterogeneous data, and the multivariate heterogeneous data comprises satellite remote sensing, unmanned plane oblique photography and ground mobile laser point cloud. Constructing a cross-scale courtyard enclosure data cube through the fused data, cutting the data cubes of different scales into a plurality of regions, constructing a three-dimensional space cube through the cut regions, and collecting cultural feature information related to historical courtyards through the three-dimensional space cube; according to the method, satellite remote sensing, unmanned aerial vehicle oblique photography and ground mobile laser point cloud are fused through a multi-source heterogeneous data adaptive fusion mechanism, a cross-scale courtyard enclosure data cube is constructed, and a data weighted registration algorithm based on feature significance is proposed. The problems of data missing and registration deviation caused by high building density and serious shielding in historical blocks are solved.

A method and system for detecting small and weak infrared targets in complex backgrounds with multi-feature fusion belong to the technical field of infrared target detection and recognition, and solve the problem that the existing technology has poor algorithm adaptability when detecting small and weak infrared targets in complex backgrounds, resulting in a large number of false alarms in the detection results and affecting the accuracy of the detection results. The present invention first extracts radiation features, multi-order directional derivative features and spectral features representing the radiation characteristics, structural characteristics and intensity characteristics of small and weak targets respectively, fuses multiple features to construct a feature saliency map, enhances the target while suppressing background noise; then uses the CFAR adaptive detection method to calculate the segmentation threshold of the image, obtains a binary segmentation result, and performs morphological processing to screen out false targets caused by isolated points and noise, and obtains the final detection result of small and weak targets. The algorithm of the present invention has low complexity, strong adaptability to complex backgrounds, high detection accuracy and is convenient for engineering implementation.

The invention relates to the technical field of gift box package appearance detection, in particular to an efficient flaw detection system for batch production of gift box packages based on computer vision. The system comprises a texture direction analysis module which is used for collecting a surface image of a gift box package, preprocessing the surface image to obtain a surface gray level image, and obtaining direction consistency of each pixel point; the suspected crease area acquisition module is used for acquiring a linear feature saliency map and acquiring a suspected crease area; the crease suspicion index acquisition module is used for acquiring a crease suspicion index of the suspected crease area; the crease anomaly detection module is used for acquiring a comprehensive anomaly index of a suspected crease region according to the crease suspicion index of the suspected crease region and a mean value of direction consistency of all pixel points in the suspected crease region; and judging whether the suspected crease region is a crease flaw region or not according to the comprehensive abnormal index. According to the invention, the accuracy of gift box package crease detection can be improved.

The invention relates to the field of computer image processing, in particular to a millimeter-level tea pest detection method and device based on improved YOLOv8. According to the method provided by the invention, an existing method for detecting pests by utilizing YOLOv8 is improved, and the feature significance of millimeter-level tea pests in a target detection method based on machine learning is effectively improved by adopting a slice-assisted reasoning SAHI strategy; gating convolution additive self-attention (GCAS) is adopted, the capacity of an existing network for capturing key information of small target tea pests is enhanced, and the detection precision is further improved. Aiming at the special problems of insect state postures, shielding and the like in the field of tea pest detection, the fine-grained feature extraction capability of multiple small target tea pests of a network is improved by using space depth conversion convolution (SPD)-Conv, and all-kernel modules OKM of different receptive fields are adopted to respectively pay attention to tea pest features of different scales; effective detection of pests with different postures and overlapped sheltered pests is realized.

The invention relates to an X-ray image quality evaluation method and device, computer equipment and a storage medium. Degradation factor simulation operation and degradation factor combination operation are carried out on original X-ray images, degraded X-ray images corresponding to each degradation factor combination mode and category labels corresponding to the degraded X-ray images are obtained to form a training image set, and category label information of the degraded X-ray images can be reflected. According to the method, features of multiple convolution stages are obtained through processing in a convolutional neural network model, feature saliency learning is carried out, multi-scale saliency features are obtained, local, channel and global features can be better captured, category label information of each distorted image can be identified through the trained convolutional neural network model, and the classification label information of each distorted image can be identified through the trained convolutional neural network model. Therefore, the method is suitable for monitoring the image transmission quality of the security check system, intelligently evaluating the quality and querying the image degradation reason, reduces the participation of technicians to a certain extent, and improves the error correction capability of distorted images.

The invention discloses a data acquisition pre-filtering processing system based on three-dimensional laser scanning. The data acquisition pre-filtering processing system comprises a multi-modal data acquisition module which is used for synchronously acquiring geometric characteristics, multispectral intensity data, polarization parameters and positioning and attitude determination information; the multi-modal vegetation penetration preprocessing module is used for outputting pure terrain point cloud and vegetation masks; the terrain self-adaptive dynamic parameter optimization module is used for acquiring pure terrain point clouds as source point clouds and target point clouds, searching candidate corresponding points in the target point clouds for each point in the source point clouds, performing multiple filtering on the candidate corresponding point pairs, performing coordinate transformation on the source point clouds by applying the optimal rigid body transformation matrix, and performing dynamic parameter optimization on the target point clouds; calculating a registration error between the transformed point cloud and the target point cloud; the micro-variation feature enhancement multi-scale analysis module is used for receiving the dynamically optimized parameters and carrying out multi-scale pyramid decomposition and feature saliency detection on the pure terrain point cloud; and the quality evaluation and output module is used for carrying out quality evaluation on the processed point cloud.

The invention relates to the technical field of gas detection, in particular to a gas detection method and system based on weak signal enhancement and time domain denoising. The method comprises the steps of constructing an input image set based on a detection image; preprocessing each frame of image in the input image set; based on the gray values of any frame of image and the next frame of image, obtaining the gray value variation at each pixel point of the any frame of image, and constructing a variation vector; determining a threshold value based on the change vector, setting the gray value of the pixel point of which the gray value variation is lower than the threshold value as 0, and determining the gray value of the pixel point of which the gray value variation is lower than the threshold value; and carrying out post-processing on any frame of image to obtain an output image set and outputting the output image set. According to the invention, the feature saliency of the small-flow gas in the image can be effectively improved.

The invention relates to the technical field of image recognition, in particular to a silkworm disease recognition system which comprises an image gray analysis module, a self-adaptive image enhancement module, a multi-dimensional feature extraction module, a feature saliency evaluation module, a dynamic weight distribution module and a disease classification mapping module. Identifying an overall gray level distribution condition and a local gray level fluctuation condition in the original image data of the target silkworm body to determine a dynamic adjustment parameter, and correcting the original image data to obtain enhanced image data; integrating multi-dimensional feature information in the enhanced image data into silkworm body state feature information; judging the significance degree of different types of information in the silkworm body state characteristic information for distinguishing different diseases so as to allocate the emphasis proportion of the characteristic information in the silkworm body state characteristic information; fusing the emphasis proportion and the silkworm body state feature information, and mapping the fused silkworm body state feature information to a preset silkworm disease feature template library to obtain disease categories; according to the invention, the accuracy of silkworm disease recognition can be improved.

The invention relates to the technical field of food package detection, in particular to a ready-to-eat prefabricated dish package airtightness identification method based on image analysis, which comprises the following steps: calculating a clear quality quantitative index through an infrared image collected in real time to obtain a high-quality package image; extracting multi-modal defect features such as visible light edge, infrared temperature and texture from the packaging image, constructing a feature saliency matrix, and performing multi-modal fusion compensation based on the reliability of each feature channel to obtain a standard feature saliency matrix; calculating a defect confidence coefficient according to the matrix so as to determine whether to trigger air tightness physical reinspection or not; and finally, combining the defect confidence coefficient with a physical recheck result, calculating an air tightness index, and completing the final judgment of the air tightness of the package. Through the multi-modal feature fusion and reliability compensation mechanism, the accuracy, the robustness and the automation level of package airtightness recognition are remarkably improved, and the defects that in the prior art, dependence on a single feature is high, and misjudgment is prone to occurring in a complex environment are effectively overcome.

The invention relates to the technical field of data processing, in particular to a multi-modal data processing method and system based on dynamic weight fusion, and the method comprises the steps: obtaining multi-modal data, and carrying out the standardization preprocessing; constructing a dynamic weight model based on an attention mechanism, constructing a cross-domain knowledge graph, and performing weighted fusion on the multi-modal features according to a dynamic weight coefficient matrix; establishing a real-time feedback mechanism, and iteratively updating the model by using parameters after feedback optimization; the method has the advantages that the dynamic weight of each modal is calculated in real time through the attention mechanism, and compared with an existing fixed weight method, the data processing efficiency in a complex scene is improved by more than 30%. In a multimedia public opinion analysis scene, the weight can be dynamically adjusted according to text emotion intensity, image visual feature saliency and audio intonation change, and public opinion homes are accurately captured.

The application discloses an adversarial sample visual explanation method based on a class activation map, and comprises an adversarial sample generation stage, an adversarial sample corresponding to a normal sample is generated by adding disturbance to the normal sample through an adversarial attack algorithm; a feature extraction stage, an adversarial sample feature and a normal sample feature are obtained by inputting an adversarial sample picture and a normal sample picture into a trained deep learning model for feature extraction, and sample classification is performed on the adversarial sample and the normal sample; a feature saliency map generation stage, a gradient of a classification result of the adversarial sample to the adversarial sample feature and a gradient of a classification result of the normal sample to the normal sample feature are calculated; and the gradient difference obtained by calculating the normal sample and the adversarial sample is taken as a weight, and linearly weighted fusion is performed on the adversarial sample feature to obtain a final adversarial sample feature saliency map.

The application provides a kind of UWF fundus image lesion area positioning method based on weakly supervised learning, comprising: obtaining UWF fundus image;ResNet convolutional neural network model is constructed;Obtain multi-scale feature data;Obtain global attention feature data;The multi-scale feature data and global attention feature data are added to obtain feature saliency data;Filtering;Positioning of lesion area;Each of the positioned data is marked;Probabilistic mapping is carried out on the data with marks;Classifier is constructed, and the mapped data and the data with marks are trained;The lesion area in test image set is positioned and predicted by the trained classifier;The application only needs to carry out simple classification label to mark the model, form weakly supervised learning, and can effectively position the lesion area at last, and artificial marking is not needed on the scale of fundus image pixel level, and the artificial cost is greatly reduced.

The invention relates to the technical field of material detection, and discloses an accurate target positioning method and system for flaw detection, and the method comprises the steps: synchronously collecting multi-source flaw detection signals, carrying out the preprocessing of the multi-source flaw detection signals, obtaining a standardized signal sequence, carrying out the fusion analysis of the standardized signal sequence, and generating a fusion feature map; recognizing a suspected defect area according to the fused feature map to obtain a preliminary defect positioning coordinate set, sorting the preliminary defect positioning coordinate set based on the coordinate feature saliency of the fused feature map, generating a priority scanning path, and calculating the signal confidence of the priority scanning path in real time, the priority scanning path is dynamically adjusted based on a preset self-adaptive threshold value to obtain a dynamic scanning path, and multi-angle cross validation is performed on a scanning result of the dynamic scanning path to obtain a final positioning result; according to the method, through feature-level dynamic fusion of the multi-source signals, the recognition sensitivity and positioning precision of complex defects are remarkably improved.

The present application relates to a method, device and equipment for adaptive redundancy removal of scattering centers based on feature saliency. The method includes: extracting multi-layer feature maps of an ISAR target image through a target recognition convolutional neural network, calculating the contribution scores of each layer of feature maps to the prediction and classification results of the target recognition convolutional neural network, using the contribution scores of each layer of feature maps as weights, performing point-by-point multiplication on each layer of feature maps to obtain a class activation heat map, precisely matching the class activation heat map and the target original model in the ISAR target image through a calibration method, calculating the feature saliency values of each scattering center based on the matched class activation heat map, and dynamically selecting the scattering centers with significant features according to the feature saliency values of each scattering center to achieve adaptive redundancy removal of scattering centers. By using this method, the scattering centers in the ISAR image can be efficiently redundant removed, and the key feature information can be ensured to be retained in the image after redundancy removal.

The invention provides a simulated sonar image post-processing fusion system, which comprises the following functional modules: a saliency evaluation module used for preprocessing a plurality of simulated sonar images of the same scene and inputting the preprocessed simulated sonar images into a target saliency scoring model to obtain a target saliency score of each image; the feature fusion module is used for calculating an average value and a consistency index based on the score to obtain a feature significance index; the environment adaptability dynamic threshold value calculation module is used for collecting the turbidity of the water body and calculating a dynamic characteristic significance threshold value; and the target state credibility evaluation module judges the target state by comparing the index with a threshold value, and outputs a credibility grade in combination with the turbidity and the consistency index. The method realizes noise suppression and stability improvement under multi-image fusion, gives consideration to detection sensitivity and reliability in a complex underwater scene through environment adaptive threshold dynamic adjustment, outputs a target state with credibility evaluation, and enhances the intelligent level and decision support capability of the system.

This invention belongs to the field of image data processing technology, specifically relating to a method and system for detecting surface defects in fabrics based on a deep convolutional neural network. The method includes: acquiring an original fabric image and performing adaptive bilateral guided preprocessing to obtain a preprocessed fabric image; obtaining a structural breakage index for evaluating texture uniformity based on the gradient field distribution of the preprocessed fabric image; inputting the structural breakage index into a convolutional neural network to obtain a deep feature image, and using the Hessian matrix to evaluate the curvature abrupt changes in the feature distribution to obtain feature saliency; performing centroid aggregation on the feature saliency, and combining the spatial distribution and geometric shape between defect areas to obtain geometric interaction potential energy; and driving the loom to execute deceleration commands and physical markers based on the geometric interaction potential energy. This invention solves the problem of defect identification in high-density fabrics by assessing the spatial clustering risk of defects, thus achieving precise control of fabric quality.

The present application relates to the technical field of packaging detection, in particular to a test paper bag packaging detection system based on machine vision, which comprises an image acquisition module, a boundary construction module, a closure inspection module, a direction evaluation module and a quality judgment module.In the present application, the safe margin area of the seal is cut in image acquisition, the focusing of the effective image is improved, the edge contrast is strengthened by gray threshold screening, the target feature saliency is enhanced, the edge connection is constructed based on coordinate gradient, the path continuity and structure rules are ensured, the contour point column is generated in topological order, the breakpoint identification and closed state quantitative analysis are supported, the angle sequence is constructed by the direction difference of adjacent points, the direction consistency evaluation precision is improved by comparing with the standard template bit by bit, the defect positioning is output by combining the breakpoint coordinates and the direction deviation, the visualized directional marker is formed, the abnormal marker is clearly processed, and the test paper bag packaging detection fine identification is realized.

The embodiment of the invention relates to an image processing method and device, equipment and a storage medium, and the method comprises the steps: firstly obtaining a feature vector set corresponding to a target image, and the feature vector set comprises a visual feature vector extracted from the target image; then determining a feature saliency degree value corresponding to each visual feature vector in the feature vector set, wherein the feature saliency degree value is used for representing the saliency degree of the visual feature described by the visual feature vector in the target image; and then based on the feature saliency degree value, a target visual feature vector is determined from the feature vector set, so that a processing result of the target image is generated after the target visual feature vector is subsequently input to an image processing model, and the saliency degree of the visual feature vector described by the target visual feature vector in the target image meets a preset condition. According to the embodiment of the invention, the calculation efficiency of the image processing model is improved, and the influence of visual feature vector compression on the accuracy of an image processing result is reduced to a greater extent.