High-resolution image patch detection method, device, equipment, storage medium and product

Abstract

This application discloses a high-resolution image block detection method, apparatus, device, storage medium, and product, relating to the field of image processing technology. The method includes: acquiring a high-resolution original image to be detected; establishing a global spatial coordinate system; and dynamically overlapping and dividing the original image to be detected into blocks based on pixel spatial features to obtain several sub-images to be detected with local coordinates; performing background adaptive filtering on each sub-image to select regions of interest (ROIs); performing data transmission optimization processing on the ROIs; transmitting the ROIs to a graphics processor (GPU) for image defect detection; and performing coordinate remapping and result fusion processing on the sub-block detection results output by the GPU to obtain the global defect detection result corresponding to the original image to be detected. This application improves the overall efficiency and real-time performance of high-resolution image detection, directly matching the actual application needs of industrial inspection and enhancing the practicality of the detection results.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a high-resolution image block detection method, apparatus, device, storage medium, and product. Background Technology

[0002] In the field of high-resolution industrial visual inspection (25 megapixels and above), the industry has gradually adopted tiled detection technology to address the challenges of processing high-resolution images, making it the mainstream technique in this field. Existing tiled detection technology segments high-resolution images into fixed blocks, then inputs the segmented sub-blocks into a deep learning model for inference and detection. During processing, large images need to be frequently copied between disk, memory, and video memory. Deep inference operations are then performed on all segmented sub-blocks. The overall process follows a fixed processing logic and lacks targeted optimization for image features and data transmission.

[0003] Existing technologies have significant technical shortcomings. On the one hand, the fixed block method is prone to edge cutting effects, cutting off defects located at the boundaries, resulting in incomplete features and thus causing missed detections. On the other hand, deep inference is also performed on defect-free background blocks, creating a large amount of computational redundancy and causing serious waste of computing power. At the same time, cross-media copying of large-size images takes too long, forming a significant IO bottleneck, which affects the inference efficiency of the system and keeps the throughput at a low level. These problems together make it difficult for the overall effect and processing efficiency of high-resolution industrial vision inspection to meet the needs of actual industrial applications. Summary of the Invention

[0004] The main objective of this application is to provide a high-resolution image block detection method, apparatus, device, storage medium, and product, which aims to improve the overall effect and processing efficiency of high-resolution industrial visual inspection.

[0005] To achieve the above objectives, this application proposes a high-resolution image block detection method, the method comprising: A high-resolution original image to be detected is obtained, a global spatial coordinate system is established, and the original image to be detected is dynamically overlapped and divided into blocks based on pixel spatial features to obtain several sub-images to be detected with local coordinates. Background adaptive filtering is performed on each of the sub-images to be detected to select the region of interest sub-blocks; The region of interest sub-blocks are subjected to data transmission optimization processing, and the region of interest sub-blocks are transmitted to the graphics processor for image defect detection. The sub-block detection results output by the graphics processor are then subjected to coordinate remapping and result fusion processing to obtain the global defect detection results corresponding to the original image to be detected.

[0006] In one possible implementation, the original image to be detected is dynamically overlapped and segmented based on pixel spatial features to obtain several sub-images to be detected with local coordinates, including: Obtain preset defect feature parameters and image gradient parameters to determine the dynamic overlap area; Edge feature perception is performed on the original image to be detected to identify high gradient regions in the original image to be detected; The block step size is adjusted according to the dynamic overlap region and the high gradient region, and the original image to be detected is divided into blocks to obtain several sub-images to be detected with local coordinates.

[0007] In one possible implementation, the step of performing background adaptive filtering on each of the sub-images to be detected to select the region of interest sub-blocks includes: Lightweight feature analysis is performed on each of the sub-images to be detected to obtain the statistical features and edge features of the sub-images to be detected; Obtain a preset feature threshold, and compare the statistical features and the edge features with the preset feature threshold respectively; Based on the comparison results, a branch decision is made on the sub-image to be detected to select the region of interest sub-blocks that meet the feature requirements.

[0008] In one possible implementation, the step of performing data transmission optimization processing on the sub-block of the region of interest, transmitting the sub-block of the region of interest to the graphics processor for image defect detection, and performing coordinate remapping and result fusion processing on the sub-block detection results output by the graphics processor to obtain the global defect detection result corresponding to the original image to be detected includes: The region of interest (ROI) sub-block is processed using multi-buffer and asynchronous bus transmission, and then transmitted to the graphics processor for image defect detection. The sub-block detection result output by the graphics processor is obtained. For any of the sub-block detection results, coordinate remapping is performed on the sub-block detection results according to the global spatial coordinate system to restore the global coordinates of the sub-block detection results; The sub-block detection results under the global coordinates are fused to obtain the global defect detection results corresponding to the original image to be detected.

[0009] In one possible implementation, the step of performing multi-buffer and asynchronous bus transmission processing on the region of interest sub-block and transmitting the region of interest sub-block to the graphics processor for image defect detection includes: Independent write and read buffers are allocated in memory to enable parallel execution of image acquisition and preprocessing. The region of interest (ROI) sub-block is copied to the graphics processor using asynchronous transmission technology. Where hardware supports it, a unified storage architecture is used to reduce data copying operations between different storage media. The ROI sub-block is then transmitted to the graphics processor for image defect detection.

[0010] In one possible implementation, the fusion processing of the sub-block detection results under the global coordinates to obtain the global defect detection result corresponding to the original image to be detected includes: Based on the global coordinates, identify duplicate detection results located in overlapping areas of the sub-image to be detected; Obtain the position weighting rules for the repeated detection results, and assign corresponding edge response weights to each repeated detection result; The non-maximum suppression algorithm, combined with the edge response weights, is used to filter and integrate the repeated detection results to obtain the global defect detection results corresponding to the original image to be detected.

[0011] In one possible implementation, the method is applied to an online testing scenario for waterproof and breathable membranes, and the method further includes: A high-resolution original image of the waterproof and breathable membrane to be inspected is acquired during the high-speed movement of the roll material. A global spatial coordinate system is established, and the original image to be inspected is dynamically overlapped and divided into blocks based on pixel spatial features to obtain several sub-images to be inspected with local coordinates. Background adaptive filtering is performed on each of the sub-images to be detected to remove defect-free background sub-blocks and filter out the regions of interest sub-blocks containing potential defects; The region of interest (ROI) sub-blocks are processed for data transmission optimization, and then transmitted to the graphics processor. The graphics processor detects and identifies micron-sized holes and elongated folds within the ROI sub-blocks, and outputs sub-block detection results with local coordinates. The sub-block detection results are subjected to coordinate remapping and result fusion processing to restore the local coordinates of micron-sized holes and elongated folds to the global spatial coordinate system, thereby obtaining the global defect detection results corresponding to the original image to be detected.

[0012] Furthermore, to achieve the above objectives, this application also proposes a high-resolution image block detection device, which includes: The acquisition module is used to acquire a high-resolution original image to be detected, establish a global spatial coordinate system, and perform dynamic overlapping and segmentation of the original image to be detected based on pixel spatial features to obtain several sub-images to be detected with local coordinates. The filtering module is used to perform background adaptive filtering on each of the sub-images to be detected, and to filter out the sub-blocks of the region of interest. The detection module is used to perform data transmission optimization processing on the sub-blocks of the region of interest, transmit the sub-blocks of the region of interest to the graphics processor for image defect detection, and perform coordinate remapping and result fusion processing on the sub-block detection results output by the graphics processor to obtain the global defect detection results corresponding to the original image to be detected.

[0013] In addition, to achieve the above objectives, this application also proposes a high-resolution image block detection device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the high-resolution image block detection method as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the high-resolution image block detection method described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the high-resolution image block detection method described above.

[0016] This application provides a high-resolution image block detection method, apparatus, device, storage medium, and product. The high-resolution image block detection method establishes a global spatial coordinate system for the high-resolution original image to be detected and performs dynamic overlapping block segmentation based on pixel spatial features. This effectively avoids defects being cut off, ensures the integrity of defect features, and reduces the probability of missed detection. Furthermore, by performing background adaptive filtering on the segmented sub-images to be detected, defect-free background blocks can be removed in advance, reducing invalid inference by the graphics processor and significantly reducing computing power consumption. By optimizing data transmission for sub-blocks in the region of interest, data transmission efficiency can be improved, IO bottlenecks can be eliminated, and the overall system throughput can be increased. At the same time, coordinate remapping and result fusion are performed on the sub-block detection results output by the graphics processor. This can restore the global position of defects in the original image and solve the problem of repeated detection in overlapping areas, improving the accuracy of detection results. This not only improves the overall efficiency and real-time performance of high-resolution image detection but also outputs global defect detection information of the original image, directly matching the actual application needs of industrial inspection and improving the practicality of the detection results. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating an embodiment of the high-resolution image block detection method of this application. Figure 2 A schematic diagram of the background filtering process provided for the high-resolution image block detection method of this application; Figure 3 A timing diagram for the asynchronous scheduling of a multi-level task pipeline provided for the high-resolution image block detection method of this application; Figure 4 A flowchart illustrating the inverse coordinate mapping process provided for the high-resolution image block detection method of this application; Figure 5 A flowchart illustrating the NMS result fusion process for the high-resolution image block detection method of this application; Figure 6 This is a schematic diagram of the device structure of the hardware operating environment involved in the high-resolution image block detection method in this application embodiment.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device, big data service platform, or high-resolution image block detection system capable of realizing the above functions. The following description uses a high-resolution image block detection system as an example to illustrate this embodiment and the subsequent embodiments.

[0024] Based on this, embodiments of this application provide a high-resolution image block detection method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the high-resolution image block detection method of this application.

[0025] In this embodiment, the high-resolution image block detection method includes steps S11 to S13: Step S11: Obtain a high-resolution original image to be detected, establish a global spatial coordinate system, and perform dynamic overlapping and segmentation of the original image to be detected based on pixel spatial features to obtain several sub-images to be detected with local coordinates. It should be noted that a high-resolution original image to be inspected refers to the original image whose pixel scale meets the high requirements of industrial visual inspection and is required for defect detection. The global spatial coordinate system refers to the overall coordinate system established for the original image to be inspected, which can pinpoint the position of any pixel in the image. Pixel spatial features refer to the gradients, edges, and other features of the original image to be inspected that reflect the image content in the pixel dimension. Dynamic overlapping segmentation refers to the image segmentation method that adjusts the segmentation method according to image features and ensures overlapping areas between adjacent sub-images. The dynamic overlap compensation mechanism improves the detection recall near the cutting line and ensures the physical connectivity of continuous defects. Local coordinates refer to the independent coordinate system within the sub-image to be inspected. The sub-image to be inspected refers to several small image blocks obtained after segmentation of the original image to be inspected.

[0026] Furthermore, this step aims to address the problem of missed detections due to defects caused by fixed segmentation. By establishing a unified global spatial coordinate system for the original image to be detected, the coordinate tracing of all subsequent detection results has a unified standard. At the same time, dynamic overlapping segmentation is performed in combination with pixel spatial features, which can detect potential defect areas in the image. By adjusting the segmentation step size to set the overlapping area, it is ensured that the defect is fully presented in at least one sub-image to be detected, thus avoiding the edge cutting effect from the root.

[0027] Specifically, in one possible implementation, the system can directly acquire high-resolution images of the industrial production scene using an image acquisition device as the original image to be detected, or it can read previously acquired high-resolution industrial inspection images from a storage medium as the original image to be detected. See steps S21-S23 for details, which will not be repeated here.

[0028] Step S12: Perform background adaptive filtering on each of the sub-images to be detected to select the sub-blocks of the region of interest; It should be noted that background adaptive filtering refers to a processing method that automatically determines whether a sub-image to be detected has a defect-free background based on its image features and performs targeted filtering. In one embodiment, for a 25-megapixel resolution image, background filtering and parallel scheduling stabilize the end-to-end closed-loop detection time within 300 milliseconds. A region of interest (ROI) sub-block refers to an image sub-block in the sub-image to be detected that contains potential defect features and requires further deep defect detection. The purpose of this operation is to eliminate defect-free background sub-blocks, reducing invalid inference by the graphics processor and lowering computational power consumption. In one possible implementation, the system can perform background adaptive filtering in the central processing unit (CPU), utilizing the CPU's computing power to complete lightweight feature analysis and avoiding the consumption of GPU computing resources.

[0029] Specifically, please refer to steps S31-S33 and Figure 2 This will not be elaborated upon here.

[0030] Step S13: Perform data transmission optimization processing on the sub-block of the region of interest, transmit the sub-block of the region of interest to the graphics processor for image defect detection, and perform coordinate remapping and result fusion processing on the sub-block detection results output by the graphics processor to obtain the global defect detection result corresponding to the original image to be detected.

[0031] It should be noted that data transmission optimization refers to improving the efficiency of transmitting region-of-interest (ROI) sub-blocks from the central processing unit (CPU) to the graphics processing unit (GPU) by optimizing the storage buffer and data transmission method. The GPU refers to a hardware device with high parallel computing power capable of performing deep image defect detection inference. Image defect detection refers to the process of using a deep learning model to process RIO sub-blocks, identify the type and location of defects, and output detection results. Sub-block detection results refer to the defect information output by the GPU after detecting defects in a single RIO sub-block, based on the local coordinates of the sub-image. Coordinate remapping refers to the process of converting the sub-block detection results based on the local coordinates of the sub-image to be detected into coordinates based on the global spatial coordinate system of the original image to be detected. Result fusion refers to the process of integrating all sub-block detection results after coordinate remapping, eliminating duplicate detection results in overlapping areas, and obtaining a unified and accurate detection result. Global defect detection results refer to the final detection results based on the global spatial coordinate system of the original image to be detected, which can reflect the type, location, size, and other information of all defects in the original image.

[0032] Furthermore, the purpose of this system operation is to solve the IO bottleneck problem of high-resolution image data transmission, while achieving accurate restoration and integration of defect detection results. The implementation involves three stages: the first stage optimizes data transmission through multiple buffers and asynchronous bus transmission, enabling parallel execution of image acquisition, preprocessing, and data transmission, improving transmission efficiency and eliminating the IO bottleneck; the second stage restores the sub-block detection results in local coordinates output by the graphics processor to global coordinates based on the global spatial coordinate system, ensuring that the position of each defect corresponds to the actual position in the original image to be detected; the third stage fuses the detection results in global coordinates, eliminating duplicate detection results in overlapping areas to improve the accuracy of the detection results, ultimately obtaining the global defect detection results of the original image to be detected. This utilizes a dynamic overlap strategy to compensate for the cutting boundaries in pixel space, and, in conjunction with a three-stage pipeline asynchronous scheduling, solves the problems of high-resolution images being "unable to transmit, slow to process, and incomprehensible."

[0033] Specifically, please refer to steps S41-S63 and Figure 3 , Figure 4 , Figure 5 This will not be elaborated upon here.

[0034] This embodiment first establishes a global spatial coordinate system for the high-resolution original image to be inspected and performs dynamic overlapping and segmentation based on pixel spatial features, ensuring the integrity of defect features and reducing the probability of missed detections from the root. Then, it uses adaptive background filtering to remove defect-free background sub-blocks, reducing the ineffective computing power consumption of the graphics processor and improving computing power utilization efficiency. Finally, it solves the IO bottleneck problem of high-resolution image data transmission through data transmission optimization, and restores the detection results of the graphics processor to the global coordinates through coordinate remapping. By fusing the results and removing duplicate detection results, it achieves accurate restoration and integration of defect detection results, and finally obtains the global defect detection results of the original image to be inspected. It fully leverages the hardware advantages of the lightweight processing of the central processing unit and the high parallel inference of the graphics processor, effectively solving the problems of edge cutting, computing power waste, and IO bottleneck in existing high-resolution industrial vision inspection, and improving the real-time performance, accuracy, and computing power utilization efficiency of the inspection.

[0035] In one feasible implementation, the step of dynamically overlapping and dividing the original image to be detected based on pixel spatial features to obtain several sub-images to be detected with local coordinates includes: Step S21: Obtain preset defect feature parameters and image gradient parameters to determine the dynamic overlap area; It should be noted that the preset defect feature parameters refer to parameters that are pre-set according to the defect recognition requirements of industrial vision inspection and can characterize the basic features of the defect to be detected, such as the minimum defect major diameter. The image gradient parameters refer to parameters that reflect the rate of change of pixel grayscale values ​​in the original image to be detected. The dynamic overlap area refers to the image region that overlaps between adjacent sub-images to be detected, which is dynamically determined according to the preset defect feature parameters and the image gradient parameters. The purpose of the system performing this operation is to set a reasonable overlap area range for the dynamic overlap segmentation, so that the overlap area can meet the requirement of fully presenting potential defects and avoid defects being cut off due to segmentation.

[0036] Specifically, the system reads the preset defect feature parameters for the current industrial inspection scenario from the local configuration file. Simultaneously, it performs pixel-level gradient calculations on the high-resolution original image to be inspected, extracting the corresponding image gradient parameters. Then, the system combines the numerical characteristics of these two types of parameters to comprehensively determine the required overlapping area range. This determines the dynamic overlapping area needed for segmenting the original image to be inspected, ensuring that this dynamic overlapping area provides space for the complete presentation of potential defects. For example, the dynamic overlapping area is set to $O = max(D min ,σ), where D min σ is the preset minimum defect length diameter, and σ is the scale of local structural changes in the image. That is, the overlap width of the slices must simultaneously satisfy the following conditions: it can completely cover the minimum defect + it can cover the edge change area.

[0037] Step S22: Perform edge feature perception on the original image to be detected and identify high gradient regions in the original image to be detected; It should be noted that edge feature perception refers to the process by which the system analyzes the pixel features of the original image to be detected and identifies the regions in the image that have edge features. High gradient regions refer to image regions in the original image to be detected where the pixel grayscale values ​​change drastically and are likely to be potential defect edges.

[0038] Specifically, the system uses an adapted visual detection algorithm to perform a full-image scan of the high-resolution original image to be detected, compares and analyzes the grayscale values ​​of each pixel and its surrounding pixels, captures pixel regions where the grayscale values ​​change significantly, integrates and marks these pixel regions as high gradient regions in the original image to be detected, and records the position information of the high gradient region in the global spatial coordinate system to provide a position reference for subsequent block step size adjustment.

[0039] Step S23: Adjust the block step size according to the dynamic overlap area and the high gradient region, and perform block processing on the original image to be detected to obtain several sub-images to be detected with local coordinates.

[0040] It should be noted that the block step size refers to the pixel movement distance between adjacent sub-images when the system divides the original image to be detected into blocks. Block processing refers to the process by which the system segments the original image to be detected according to the adjusted block step size, combined with dynamic overlapping areas. The purpose of this operation is to ensure that high-gradient regions fall completely within a specific sub-image to be detected by dynamically adjusting the block step size, thereby completely eliminating edge-cutting effects. Simultaneously, each segmented sub-image is assigned local coordinates to facilitate coordinate tracing of subsequent detection results.

[0041] Specifically, the system first plans the initial block path and initial block step size of the original image to be detected. It then compares the identified high-gradient region location information with the initial block path. When it detects that the initial tangent line will pass through a high-gradient region, it adjusts the block step size according to the determined dynamic overlap area to avoid the complete rendering range of the high-gradient region, ensuring that the high-gradient region falls completely within a single sub-image to be detected. Subsequently, the system uniformly divides the entire original image to be detected into blocks according to the adjusted block step size and the dynamic overlap area. After block processing, an independent local spatial coordinate system is established for each segmented image block, and corresponding local coordinates are assigned, ultimately resulting in several sub-images to be detected with local coordinates. Therefore, when the system detects that the tangent line passes through a high-gradient region (potential defect edge), it automatically adjusts the sub-block step size to ensure that at least one of the adjacent sub-blocks has a complete defect projection.

[0042] This embodiment first obtains preset defect feature parameters and image gradient parameters to determine the dynamic overlap area, setting an overlap range suitable for defect detection for the block segmentation operation. Then, it accurately identifies high gradient regions in the original image to be detected through edge feature perception, locating the potential defect edge positions. Finally, it dynamically adjusts the block segmentation step size by combining the dynamic overlap area and the high gradient region and completes the block segmentation process, so that the potential defect area can be completely presented in the sub-image to be detected. This completely solves the edge cutting effect problem caused by fixed block segmentation and effectively avoids the risk of missed detection due to defect cutting.

[0043] In one feasible implementation, the step of performing background adaptive filtering on each of the sub-images to be detected to select the region of interest sub-blocks includes: Step S31: Perform lightweight feature analysis on each of the sub-images to be detected to obtain the statistical features and edge features of the sub-images to be detected; It should be noted that lightweight feature analysis refers to an analysis method that extracts features from the sub-image to be detected using a computationally efficient and fast processing speed. Statistical features refer to feature information obtained through statistical calculations that reflects the overall gray-level distribution and texture complexity of the sub-image to be detected, such as local information entropy or spatial standard deviation. Edge features refer to feature information obtained through edge detection that reflects the distribution density of edge points in the sub-image to be detected, such as edge point density. The purpose of this operation is to quickly extract key features that can distinguish between background areas and potential defect areas without consuming a large amount of computing resources.

[0044] Specifically, the system performs lightweight computational processing on the CPU for each sub-graph to be detected, and calculates the statistical characteristics of the sub-graph to be detected using statistical methods, such as calculating the local information entropy or spatial standard deviation of the sub-graph to be detected. If StdDev(Tile) < τ (preset threshold), it is determined to be a pure background. Here, Tile refers to a sub-graph patch cropped from the large image, and StdDev(Tile) refers to the standard deviation of the sub-graph.

[0045] Furthermore, the system scans the sub-image to be detected using edge detection methods (such as the Sobel operator) and statistically obtains the edge features (such as edge point density) of the sub-image to be detected. The system then associates and stores the obtained statistical features with the edge features as the basis for subsequent judgment on whether the sub-image to be detected is a valid detection region.

[0046] Step S32: Obtain a preset feature threshold, and compare the statistical features and the edge features with the preset feature threshold respectively; It should be noted that the preset feature threshold refers to the numerical value of the judgment standard pre-configured by the system to distinguish between defect-free background sub-images and sub-images containing potential defects. In one possible implementation, the system sets different preset feature threshold values ​​according to different industrial inspection objects to adapt to inspection requirements. At the same time, the system can compare statistical features and edge features with independent preset feature thresholds.

[0047] Specifically, the system obtains the preset feature thresholds applicable to the current detection scenario from the configuration information, compares the statistical features of each sub-image to be detected obtained in step S31 with the corresponding preset feature thresholds, and compares the edge features of each sub-image to be detected with the corresponding preset feature thresholds. The system records the comparison results of the two features of each sub-image to be detected for subsequent branch judgment processing.

[0048] Step S33: Based on the comparison results, perform branch decision on the sub-image to be detected and filter out the region of interest sub-blocks that meet the feature requirements.

[0049] It should be noted that branch decision refers to a judgment method that divides the sub-image to be detected into different types and executes different processing logics based on feature comparison results. Regions of interest (ROIs) refer to sub-images whose feature values ​​meet preset standards, contain potential defects, and need to enter the subsequent graphics processor detection process. The purpose of this operation is to automatically remove pure background sub-images with no detection value, retaining only valid sub-blocks that require fine-grained detection, reducing subsequent unnecessary computational power consumption, and improving overall detection efficiency. In one possible implementation, the system only retains coordinate placeholder information for sub-images determined to be background, without further transmission or detection.

[0050] Specifically, the system performs branch judgment based on the comparison results of statistical features and edge features of each sub-image to be detected. When the statistical features or edge features of the sub-image to be detected meet the preset feature threshold requirements, the sub-image to be detected is marked as a region that meets the feature requirements. When neither feature of the sub-image to be detected meets the requirements, the sub-image to be detected is marked as a defect-free background region. The system filters out all sub-images marked as meeting the feature requirements to form the final region of interest (ROI) sub-blocks. That is, the discard branch: determined as meaningless background, directly skips GPU transmission and only retains coordinate placeholders; the selected branch: determined as ROI (region of interest), and pushes it into the high-performance asynchronous transmission queue. Since the model only processes ROI sub-blocks containing defects, the computing power requirement of the GPU is reduced by more than 60%, allowing deployment on lower-cost edge devices.

[0051] This embodiment first performs lightweight feature analysis on the sub-image to be detected to obtain statistical and edge features, achieving rapid feature extraction with low computational consumption. Then, by obtaining preset feature thresholds and completing feature comparison, a standardized basis for region judgment is established. Finally, based on the comparison results, branch decisions are executed to accurately remove defect-free background sub-images and select regions of interest sub-blocks. This effectively reduces the invalid inference tasks of the subsequent graphics processor, lowers the overall computational consumption of the system, improves the processing speed and resource utilization of high-resolution image defect detection, and provides an accurate data foundation for subsequent efficient parallel detection processes.

[0052] In one feasible implementation, the step of performing data transmission optimization processing on the region of interest sub-blocks, transmitting the region of interest sub-blocks to a graphics processor for image defect detection, and performing coordinate remapping and result fusion processing on the sub-block detection results output by the graphics processor to obtain the global defect detection results corresponding to the original image to be detected includes: Step S41: Perform multi-buffer and asynchronous bus transmission processing on the region of interest sub-block, transmit the region of interest sub-block to the graphics processor for image defect detection, and obtain the sub-block detection result output by the graphics processor; It should be noted that multi-buffer and asynchronous bus transmission processing refers to a processing method that achieves parallel execution of data storage, transmission, and hardware computation by opening multiple independent storage buffers and employing asynchronous transmission technology. The graphics processing unit (GPU) refers to a hardware device with high parallel computing capabilities, capable of running deep learning models to perform accurate image defect detection. Image defect detection refers to the process by which the GPU uses a preset detection model to analyze sub-blocks within the region of interest (ROI) and identify the type and location of defects. The sub-block detection result refers to the defect-related information output by the GPU after completing the detection, based on the local coordinates of the sub-image to be detected. The purpose of this system operation is to solve the IO bottleneck problem of high-resolution image data transmission, improve data transmission efficiency, and allow the GPU to focus on depth detection of sub-blocks within the ROI, ensuring detection accuracy. In one possible implementation, the system can open independent write and read buffers in memory for writing and reading image data, respectively. With hardware support, the system adopts a unified storage architecture to reduce data copying operations between different storage media.

[0053] Specifically, the system performs multi-buffer and asynchronous bus transmission processing on the selected region of interest (ROI) sub-blocks. Independent storage buffers are allocated in memory to achieve parallel reading and writing of data. Asynchronous transmission technology is used to transmit the image data of the ROI sub-blocks to the graphics processor. After receiving the data, the graphics processor runs a preset deep learning detection model to perform image defect detection on each ROI sub-block, identify the defect information within the sub-block, and generate sub-block detection results based on local coordinates. The system obtains the sub-block detection results corresponding to all ROI sub-blocks from the graphics processor and stores them temporarily.

[0054] Step S42: For any of the sub-block detection results, perform coordinate remapping processing on the sub-block detection results according to the global spatial coordinate system to restore the global coordinates of the sub-block detection results; It should be noted that coordinate remapping refers to the process of converting defect information based on the local coordinates of the sub-image to be detected into coordinates based on the global spatial coordinate system of the original image to be detected. Global coordinates refer to the coordinate information in the global spatial coordinate system of the original image to be detected that can indicate the actual location of the defect.

[0055] Specifically, for each temporarily stored sub-block detection result, the system retrieves the position association information of the corresponding sub-image to be detected in the global spatial coordinate system. According to the preset coordinate transformation rules, the local coordinates in the sub-block detection result are calculated and transformed, and all defect location information based on local coordinates is restored to global coordinates in the global spatial coordinate system of the original image to be detected. The sub-block detection results in global coordinates are obtained and associated with the data.

[0056] Step S43: Perform fusion processing on the sub-block detection results under the global coordinates to obtain the global defect detection results corresponding to the original image to be detected.

[0057] It should be noted that fusion processing refers to the process of integrating the detection results of all sub-blocks under global coordinates, eliminating duplicate detection results, correcting false alarms, and obtaining unified and accurate defect detection results. Global defect detection results refer to the final detection results based on the global spatial coordinate system of the original image to be detected, which can completely and accurately reflect the type, location, size, and other information of all defects in the original image. The purpose of this operation is to solve the problem of duplicate detection in overlapping areas caused by dynamic overlapping block segmentation, and to improve the accuracy and completeness of defect detection results.

[0058] Specifically, the system analyzes the sub-block detection results under all global coordinates, identifies duplicate detection results located in the overlapping area of ​​the sub-image to be detected, retrieves the preset result filtering rules, filters and removes duplicate detection results, retains valid defect detection information, and then integrates the valid detection results of all non-overlapping areas with the filtered overlapping area detection results, calibrates them according to the global spatial coordinate system of the original image to be detected, and finally obtains a global defect detection result that can completely and accurately reflect the defect situation of the original image to be detected.

[0059] This embodiment first processes the region of interest (ROI) sub-blocks using multi-buffer and asynchronous bus transmission, then transmits them to the graphics processor (GPU) for detection and acquisition of sub-block detection results. This effectively solves the I / O bottleneck of high-resolution image data transmission and improves the efficiency of data transmission and detection. Next, it performs coordinate remapping processing on the detection results of each sub-block according to the global spatial coordinate system, restoring the global coordinates of the defects and achieving accurate tracing of the local detection results to the actual location in the original image. Finally, it fuses the sub-block detection results under global coordinates, eliminating duplicate detection results and integrating effective defect information to obtain the global defect detection results of the original image to be inspected. This solves the problem of duplicate detection in overlapping areas, improving the accuracy and completeness of the detection results. Thus, it achieves a complete process from efficient transmission and accurate detection of ROI sub-blocks to coordinate restoration and optimization of detection results, fully leveraging the high parallel computing advantages of the GPU while ensuring the accuracy and completeness of the final defect detection results, providing reliable global defect data support for industrial vision inspection.

[0060] In one feasible implementation, the step of performing multi-buffer and asynchronous bus transmission processing on the region of interest sub-block and transmitting the region of interest sub-block to the graphics processor for image defect detection includes: Step S51: Create independent write buffers and read buffers in memory to achieve parallel execution of image acquisition and preprocessing; It should be noted that memory refers to high-speed storage components used for temporary storage of image data and intermediate processing results; write buffer refers to a storage area in memory specifically allocated for receiving and storing newly acquired image data; read buffer refers to a storage area in memory specifically allocated for reading acquired image data and performing preprocessing operations; image acquisition refers to the process of obtaining a high-resolution original image to be detected from an image acquisition device; preprocessing refers to preliminary processing operations such as dynamic overlapping and block division and adaptive background filtering on the acquired original image; and parallel execution refers to the simultaneous execution of image acquisition and preprocessing operations within the same time period without interference. The purpose of this system operation is to avoid resource contention and data blocking during image acquisition and data processing, thereby improving the smoothness and real-time performance of the overall processing flow.

[0061] Specifically, the system allocates two independent storage spaces in memory based on the size of the original image to be detected, serving as a write buffer and a read buffer respectively. That is, Buffer_A (write area) and Buffer_B (read area) are allocated in memory to achieve parallel processing of "camera writing" and "preprocessing engine reading". The image acquisition device continuously transmits and stores the high-resolution original image to be detected into the write buffer. The system simultaneously retrieves the stored image data from the read buffer to perform preprocessing operations such as dynamic overlapping block division and adaptive background filtering. Through the alternating reading and writing of the two buffers, the image acquisition process and the preprocessing process can run synchronously and in parallel, avoiding efficiency losses caused by process waiting.

[0062] Step S52: The region of interest sub-block is copied to the graphics processor using asynchronous transmission technology. Where hardware supports it, a unified storage architecture is used to reduce data copying operations between different storage media. The region of interest sub-block is then transmitted to the graphics processor for image defect detection.

[0063] It's important to note that asynchronous transmission technology refers to a transmission method where subsequent operations can continue without waiting for the copy to complete after a data copy command is sent. The heterogeneous parallel architecture completely eliminates system lag caused by image preprocessing, supporting continuous sampling by the camera at a higher frequency. Data copying refers to the process of copying image data of a region of interest (ROI) sub-block from memory to the graphics processor's (GPU) storage unit. Unified Memory refers to a hardware storage architecture where the CPU and GPU share the same physical memory, eliminating the need for repeated data copying. Different storage media refer to independent data storage components such as memory and GPU video memory. Image defect detection refers to the process where the GPU uses a deep learning model to identify and locate defects in the ROI sub-block. The system performs this operation to eliminate I / O bottlenecks in data transmission, shorten data waiting time, improve data transmission efficiency, and reduce hardware resource consumption. The implementation approach is to achieve synchronous data transmission and instruction execution through asynchronous transmission, and to reduce the number of data copies between different media through a unified storage architecture, maximizing data transmission efficiency.

[0064] Specifically, the system uses asynchronous transmission technology to send data copy instructions to the graphics processor, transferring the preprocessed region of interest (ROI) sub-block data from memory to the graphics processor. With hardware support for a unified storage architecture, the system directly establishes a shared access path between memory and the graphics processor, eliminating the need for repeated data copy operations. The data is directly read and used by the graphics processor. Once the data transfer is complete, the graphics processor immediately performs image defect detection on the ROI sub-block, identifies defect features within the sub-block, and outputs the detection results.

[0065] For example, the system uses CUDA Stream to achieve overlapping execution of data transmission and model inference of the previous frame. The sub-blocks of the region of interest of the waterproof and breathable membrane are copied to the graphics processor. Since the detection hardware supports a unified storage architecture, the system omits the redundant data copying steps. The graphics processor directly reads the sub-block data and performs defect detection of micron-level holes and long strip wrinkles.

[0066] This embodiment first establishes independent write and read buffers in memory, enabling image acquisition and preprocessing to be executed in parallel. This avoids process blocking and data waiting, significantly improving front-end processing efficiency. Then, asynchronous transmission technology combined with a unified storage architecture is used to efficiently transfer sub-blocks of the region of interest to the graphics processor, reducing data copying and eliminating I / O bottlenecks, ensuring data transmission speed perfectly matches hardware processing capabilities. The entire process achieves end-to-end parallelism in acquisition, preprocessing, data transmission, and defect detection, significantly improving the real-time performance and throughput of high-resolution image detection while reducing system resource consumption. This makes the entire detection method more suitable for the high-speed, continuous, and high-precision detection requirements of industrial environments.

[0067] In one feasible implementation, the step of fusing the sub-block detection results under the global coordinates to obtain the global defect detection results corresponding to the original image to be detected includes: Step S61: Based on the global coordinates, identify duplicate detection results located in the overlapping areas of the sub-image to be detected; It should be noted that global coordinates refer to the coordinate information used to pinpoint the actual location of defects in the global spatial coordinate system of the original image to be detected. Overlapping areas in the sub-images to be detected refer to the image regions that overlap between adjacent sub-images. Duplicate detection results refer to multiple defect detection results with similar coordinates formed by the same defect being detected multiple times in different sub-images. The purpose of this operation is to locate duplicate defect data caused by dynamic overlapping block segmentation, thus avoiding the situation where the same defect is labeled multiple times in the final result.

[0068] Specifically, the system iterates through all defect detection results under global coordinates, determines whether each detection result is within the overlapping area of ​​the sub-image to be detected based on its global coordinates, performs spatial location matching and comparison on the detection results located within the overlapping area, filters out multiple detection results belonging to the same defect and uniformly marks them as duplicate detection results, thus completing the accurate identification of duplicate data.

[0069] Step S62: Obtain the position weight rules of the repeated detection results and assign corresponding edge response weights to each repeated detection result; It should be noted that the position weight rule refers to the system's pre-set rule for assigning weight values ​​to the position of the defect detection result in the sub-graph to be detected, while the edge response weight refers to the weight value assigned to the detection result based on its position in the sub-graph, which is used to characterize the reliability of the detection result.

[0070] Specifically, the system obtains the position weight rules corresponding to the duplicate detection results from the configuration information, determines the specific position of each duplicate detection result in the sub-graph to be detected, calculates the corresponding edge response weight for each duplicate detection result according to the position weight rules, and associates and binds the weight with the detection result for subsequent filtering processing.

[0071] Step S63: The non-maximum suppression algorithm is used in conjunction with the edge response weight to filter and integrate the repeated detection results to obtain the global defect detection result corresponding to the original image to be detected.

[0072] It should be noted that the Non-Maximum Suppression (NMS) algorithm (based on IoU space constraints) refers to a filtering algorithm used to retain the best result from duplicate detection results and remove the rest. The filtering and integration process refers to retaining the highest-weighted valid detection result, deleting low-weighted duplicate results, and merging all valid results. The global defect detection result refers to the final defect detection result that is non-duplicative and accurate, based on the global coordinates of the original image to be detected. The purpose of this operation is to eliminate duplicate data and ensure the uniqueness and accuracy of the final detection result. The implementation idea is to use the edge response weight as the filtering criterion for the NMS algorithm, retaining the highest-weighted detection result in each group of duplicate results, deleting the remaining low-weighted results, and then integrating all valid detection results to form a non-duplicative, high-precision global defect detection result.

[0073] Specifically, the system inputs the duplicate detection results associated with edge response weights into the non-maximum suppression algorithm. Using weight magnitude as the filtering criterion, it retains the detection result with the highest weight in each group of duplicate detection results and discards the remaining duplicate detection results with lower weights. All the filtered valid detection results are integrated and uniformly calibrated, and then output according to the global spatial coordinate system to finally obtain the global defect detection results corresponding to the original image to be detected. For example, for duplicate detection boxes appearing in overlapping areas, the NMS algorithm based on IoU spatial constraints is used, and "edge response weights" are introduced, that is, higher weights are given to detection boxes closer to the center of the sub-block, solving the model false alarms caused by deformation of the segmentation edges.

[0074] This embodiment first identifies duplicate detection results within overlapping regions based on global coordinates, accurately locating redundant data caused by overlapping blocks. Then, it assigns edge response weights to duplicate results according to positional weight rules, achieving quantitative differentiation of detection result reliability. Finally, it employs a non-maximum suppression algorithm combined with weights to complete filtering and integration, effectively eliminating duplicate detection information and retaining the most accurate defect results. This entire process solves the problem of duplicate annotation caused by dynamic overlapping blocks, significantly improving the accuracy and neatness of detection results. The resulting global defect detection results are more reliable and standardized, fully meeting the practical needs of industrial visual inspection for high-precision, non-redundant results.

[0075] In one feasible implementation, the method is applied to an online testing scenario for waterproof and breathable membranes, and the method further includes: Step S71: Obtain a high-resolution original image of the waterproof and breathable membrane to be detected during the high-speed movement of the roll material, establish a global spatial coordinate system, and perform dynamic overlapping and segmentation of the original image to be detected based on pixel spatial features to obtain several sub-images to be detected with local coordinates. It should be noted that the high-speed movement of the roll material refers to the continuous high-speed transmission of the waterproof and breathable membrane on the industrial production line. The high-resolution original image of the waterproof and breathable membrane to be inspected refers to the high-definition original image of the membrane material acquired by image acquisition equipment in the high-speed movement scenario of the roll material for defect detection. The global spatial coordinate system refers to the overall coordinate system established for the original image of the waterproof and breathable membrane to be inspected, used to uniformly calibrate the pixel positions of the entire image. Pixel spatial features refer to the gradient, edge, and other feature information of the waterproof and breathable membrane image in the pixel dimension that can reflect the defect area. Dynamic overlapping block segmentation refers to the image segmentation method that adaptively adjusts the block size and overlap range according to the defect features and image gradient. Local coordinates refer to the independent coordinate system within each sub-image to be inspected. The sub-image to be inspected refers to the multiple small image blocks obtained after the original image of the waterproof and breathable membrane to be inspected is dynamically segmented.

[0076] The purpose of this operation is to adapt to the high-speed continuous inspection requirements of industrial production lines. It splits the original image of the ultra-high resolution waterproof and breathable membrane into sub-images suitable for inspection, while ensuring that defects are not cut off by the blocks. The idea is to acquire the membrane image of the high-speed moving roll in real time, establish a unified global coordinate, and then adaptively divide the image into blocks according to the defect-sensitive areas in the image, so as to ensure that all potential defects are completely preserved in the sub-image.

[0077] Specifically, the system acquires high-resolution images of the waterproof and breathable membrane transmitted during the high-speed movement of the roll material in real time using industrial image acquisition equipment. This image is used as the original image to be inspected. A global spatial coordinate system is established with the upper left corner of the image as the origin. The system performs pixel-level feature analysis on the original image to be inspected, identifies the high-gradient region where potential defects are located, determines the dynamic overlap area based on preset defect parameters and image gradient parameters, and divides the original image into blocks after adjusting the block step size. Local coordinates are established for each block of the image, and finally, several waterproof and breathable membrane sub-images with local coordinates are obtained.

[0078] Step S72: Perform background adaptive filtering on each of the sub-images to be detected, remove defect-free background sub-blocks, and filter out region of interest sub-blocks containing potential defects; It should be noted that background adaptive filtering refers to a processing method that automatically determines whether a sub-image is a pure background based on its image features and then filters it accordingly. A defect-free background sub-block refers to a pure membrane material background sub-image that does not contain any micron-sized pores, elongated wrinkles, or other defects. Potential defects refer to abnormal structures in the sub-image that may exist and affect the quality of the membrane material, such as micron-sized pores and elongated wrinkles. Regions of interest (ROI) sub-blocks refer to sub-blocks containing potential defect features that need to enter the high-precision detection process. The purpose of this operation is to quickly eliminate background data with no detection value, reduce unnecessary calculations by the graphics processor, and improve the overall processing speed in high-speed detection scenarios. The implementation involves performing lightweight feature calculations on each sub-image to be detected, using statistical and edge features to determine the presence of defect clues, retaining those that meet the criteria as ROI sub-blocks, and directly discarding those that do not.

[0079] Specifically, the system performs lightweight feature analysis on each of the waterproof and breathable membrane sub-images to be tested, extracts the statistical and edge features of each sub-image, compares the feature values ​​with preset feature thresholds, and performs branch decisions based on the comparison results. Sub-images that simultaneously or separately meet the feature conditions are identified as regions containing potential defects, while sub-images without feature anomalies are identified as defect-free background sub-blocks and directly removed. Finally, the system filters out the region of interest sub-blocks that contain only potential defects.

[0080] Step S73: Perform data transmission optimization processing on the region of interest sub-block, transmit the region of interest sub-block to the graphics processor, and have the graphics processor detect and identify micron-sized holes and elongated wrinkles in the region of interest sub-block, and output sub-block detection results with local coordinates; It should be noted that data transmission optimization refers to methods such as multi-buffering and asynchronous transmission to improve data transmission efficiency. The graphics processor (GPU) refers to a hardware processor with high parallel computing capabilities used for high-precision defect identification. Micrometer-level holes refer to perforations with extremely small diameters on the surface of the waterproof and breathable membrane; elongated wrinkles refer to elongated wrinkles on the surface of the waterproof and breathable membrane; and sub-block detection results refer to the defect location and type information output by the GPU based on the local coordinates of the sub-image. The purpose of this operation is to eliminate data transmission bottlenecks in high-speed production line scenarios, allowing the GPU to focus on high-precision defect identification and ensuring real-time detection. The implementation involves rapidly transmitting sub-blocks of the region of interest to the GPU through double buffering and asynchronous transmission. A deep learning model is then used to accurately identify holes and wrinkles, and the detection results based on local coordinates are output. In one possible implementation, the system employs a unified storage architecture to reduce the number of data copies, allowing the GPU to automatically identify micrometer-level defects using a trained defect detection model.

[0081] Specifically, the system performs multi-buffer and asynchronous bus transmission processing on the selected region of interest sub-blocks. An independent read / write buffer is allocated in memory to achieve parallel acquisition and preprocessing. Asynchronous transmission technology is used to efficiently transmit the sub-block data to the graphics processor. The graphics processor loads the pre-trained waterproof and breathable membrane defect detection model, performs inference detection on each sub-block, accurately identifies micron-sized holes and elongated wrinkles, records the local coordinates of defects in the sub-image, generates sub-block detection results with local coordinates, and outputs them.

[0082] Step S74: Perform coordinate remapping and result fusion processing on the sub-block detection results to restore the local coordinates of micron-sized holes and elongated folds to the global spatial coordinate system, thereby obtaining the global defect detection results corresponding to the original image to be detected.

[0083] It should be noted that coordinate remapping refers to the process of converting the local coordinates of defects into the global coordinates of the original image to be inspected. Result fusion processing refers to the process of eliminating duplicate detection results and integrating valid defect information. The global defect detection result refers to the final detection result based on the global coordinate system, fully labeling the location and type of all defects in the waterproof and breathable membrane. The purpose of this operation is to uniformly map the scattered sub-image detection results to the original image coordinates, eliminate the problem of duplicate detection in overlapping areas, and output complete detection results that can be directly used for industrial quality control. The implementation idea is to first convert all local coordinates into global coordinates, then identify duplicate results in overlapping areas, and complete the filtering and integration through weighting and non-maximum suppression algorithms, finally generating a global defect report.

[0084] Specifically, for each sub-block detection result, the system converts the local coordinates of micron-sized holes and elongated folds into global coordinates based on the position information of the corresponding sub-image to be detected in the global spatial coordinate system, completing coordinate remapping. The system analyzes the detection results under the global coordinates, identifies duplicate detection results in overlapping areas, assigns positional weights to duplicate results, and uses a non-maximum suppression algorithm to filter and integrate them, retaining the most accurate defect information. Finally, all valid defects are integrated according to the global coordinates to obtain the global defect detection results corresponding to the original image of the waterproof and breathable membrane to be detected.

[0085] This embodiment addresses the high-speed online inspection scenario of waterproof and breathable membranes. First, it acquires real-time images of the roll material and dynamically overlaps and segments them to ensure defects remain intact and are not cut off. Then, it uses adaptive background filtering to quickly remove invalid background sub-blocks, significantly improving inspection efficiency. Next, through optimized data transmission, the valid sub-blocks are sent to a graphics processor to achieve high-precision identification of micron-level holes and elongated folds. Finally, coordinate remapping and result fusion unify local inspection results into a global standard result. This entire method is fully adaptable to the high-speed, continuous, and high-precision inspection requirements of industrial production lines. It effectively avoids missed defects, reduces wasted computing power, and outputs standardized global inspection results, providing a stable, efficient, and reliable complete solution for automated quality inspection of waterproof and breathable membranes.

[0086] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0087] This application also provides a high-resolution image block detection device; please refer to [reference needed]. Figure 5 The high-resolution image block detection device includes: The acquisition module is used to acquire a high-resolution original image to be detected, establish a global spatial coordinate system, and perform dynamic overlapping and segmentation of the original image to be detected based on pixel spatial features to obtain several sub-images to be detected with local coordinates. The filtering module is used to perform background adaptive filtering on each of the sub-images to be detected, and to filter out the sub-blocks of the region of interest. The detection module is used to perform data transmission optimization processing on the sub-blocks of the region of interest, transmit the sub-blocks of the region of interest to the graphics processor for image defect detection, and perform coordinate remapping and result fusion processing on the sub-block detection results output by the graphics processor to obtain the global defect detection results corresponding to the original image to be detected.

[0088] The high-resolution image block detection device provided in this application, employing the high-resolution image block detection method in the above embodiments, can solve the technical problems in the background art. Compared with the prior art, the beneficial effects of the high-resolution image block detection device provided in this application are the same as the beneficial effects of the high-resolution image block detection method provided in the above embodiments, and other technical features in the high-resolution image block detection device are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0089] This application provides a high-resolution image block detection device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the high-resolution image block detection method in the above embodiment 1.

[0090] The following is for reference. Figure 6 This document illustrates a structural schematic diagram of a high-resolution image block detection device suitable for implementing embodiments of this application. The high-resolution image block detection device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The high-resolution image block detection device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0091] like Figure 6As shown, the high-resolution image patch detection device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the high-resolution image patch detection device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the high-resolution image patch detection device to communicate wirelessly or wiredly with other devices to exchange data. Although the figures show high-resolution image patch detection devices with various systems, it should be understood that implementing or having all of the systems shown is not required. More or fewer systems may be implemented alternatively.

[0092] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0093] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0094] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the high-resolution image block detection methods provided by the methods described above.

[0095] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0096] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0097] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A high-resolution image block detection method, characterized in that, include: A high-resolution original image to be detected is obtained, a global spatial coordinate system is established, and the original image to be detected is dynamically overlapped and divided into blocks based on pixel spatial features to obtain several sub-images to be detected with local coordinates. Background adaptive filtering is performed on each of the sub-images to be detected to select the region of interest sub-blocks; The region of interest sub-blocks are subjected to data transmission optimization processing, and the region of interest sub-blocks are transmitted to the graphics processor for image defect detection. The sub-block detection results output by the graphics processor are then subjected to coordinate remapping and result fusion processing to obtain the global defect detection results corresponding to the original image to be detected.

2. The high-resolution image block detection method as described in claim 1, characterized in that, The original image to be detected is dynamically overlapped and segmented based on pixel spatial features to obtain several sub-images to be detected with local coordinates, including: Obtain preset defect feature parameters and image gradient parameters to determine the dynamic overlap area; Edge feature perception is performed on the original image to be detected to identify high gradient regions in the original image to be detected; The block step size is adjusted according to the dynamic overlap region and the high gradient region, and the original image to be detected is divided into blocks to obtain several sub-images to be detected with local coordinates.

3. The high-resolution image block detection method as described in claim 1, characterized in that, The step of performing background adaptive filtering on each of the sub-images to be detected to select the region of interest sub-blocks includes: Lightweight feature analysis is performed on each of the sub-images to be detected to obtain the statistical features and edge features of the sub-images to be detected; Obtain a preset feature threshold, and compare the statistical features and the edge features with the preset feature threshold respectively; Based on the comparison results, a branch decision is made on the sub-image to be detected to select the region of interest sub-blocks that meet the feature requirements.

4. The high-resolution image block detection method as described in claim 1, characterized in that, The process of optimizing data transmission for the region of interest (ROI) sub-blocks, transmitting the ROI sub-blocks to the graphics processor (GPU) for image defect detection, and performing coordinate remapping and result fusion processing on the sub-block detection results output by the GPU to obtain the global defect detection results corresponding to the original image to be detected includes: The region of interest (ROI) sub-block is processed using multi-buffer and asynchronous bus transmission, and then transmitted to the graphics processor for image defect detection. The sub-block detection result output by the graphics processor is obtained. For any of the sub-block detection results, coordinate remapping is performed on the sub-block detection results according to the global spatial coordinate system to restore the global coordinates of the sub-block detection results; The sub-block detection results under the global coordinates are fused to obtain the global defect detection results corresponding to the original image to be detected.

5. The high-resolution image block detection method as described in claim 4, characterized in that, The step of performing multi-buffer and asynchronous bus transmission processing on the region of interest sub-block and transmitting the region of interest sub-block to the graphics processor for image defect detection includes: Independent write and read buffers are allocated in memory to enable parallel execution of image acquisition and preprocessing. The region of interest (ROI) sub-block is copied to the graphics processor using asynchronous transmission technology. Where hardware supports it, a unified storage architecture is used to reduce data copying operations between different storage media. The ROI sub-block is then transmitted to the graphics processor for image defect detection.

6. The high-resolution image block detection method as described in claim 4, characterized in that, The process of fusing the sub-block detection results under the global coordinates to obtain the global defect detection results corresponding to the original image to be detected includes: Based on the global coordinates, identify duplicate detection results located in overlapping areas of the sub-image to be detected; Obtain the position weighting rules for the repeated detection results, and assign corresponding edge response weights to each repeated detection result; The non-maximum suppression algorithm, combined with the edge response weights, is used to filter and integrate the repeated detection results to obtain the global defect detection results corresponding to the original image to be detected.

7. The high-resolution image block detection method as described in claim 1, characterized in that, The method is applied to online testing scenarios for waterproof and breathable membranes, and the method further includes: A high-resolution original image of the waterproof and breathable membrane to be inspected is acquired during the high-speed movement of the roll material. A global spatial coordinate system is established, and the original image to be inspected is dynamically overlapped and divided into blocks based on pixel spatial features to obtain several sub-images to be inspected with local coordinates. Background adaptive filtering is performed on each of the sub-images to be detected to remove defect-free background sub-blocks and filter out the regions of interest sub-blocks containing potential defects; The region of interest (ROI) sub-blocks are processed for data transmission optimization, and then transmitted to the graphics processor. The graphics processor detects and identifies micron-sized holes and elongated folds within the ROI sub-blocks, and outputs sub-block detection results with local coordinates. The sub-block detection results are subjected to coordinate remapping and result fusion processing to restore the local coordinates of micron-sized holes and elongated folds to the global spatial coordinate system, thereby obtaining the global defect detection results corresponding to the original image to be detected.

8. A high-resolution image block detection device, characterized in that, include: The acquisition module is used to acquire a high-resolution original image to be detected, establish a global spatial coordinate system, and perform dynamic overlapping and segmentation of the original image to be detected based on pixel spatial features to obtain several sub-images to be detected with local coordinates. The filtering module is used to perform background adaptive filtering on each of the sub-images to be detected, and to filter out the sub-blocks of the region of interest. The detection module is used to perform data transmission optimization processing on the sub-blocks of the region of interest, transmit the sub-blocks of the region of interest to the graphics processor for image defect detection, and perform coordinate remapping and result fusion processing on the sub-block detection results output by the graphics processor to obtain the global defect detection results corresponding to the original image to be detected.

9. A high-resolution image block detection device, characterized in that, The high-resolution image block detection device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the high-resolution image block detection method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the high-resolution image block detection method as described in any one of claims 1 to 7.

Images