An inspection image filtering method and filtering device

Abstract

The present invention provides a patrol image filtering method and a filtering device, which solve the technical problems of low efficiency of existing image quality inspection and incomplete coverage of quality inspection. The method includes: forming a feature extraction service to extract and quantify features of an image to form image feature data; forming a gallery storage service to establish a storage structure and an index structure to respectively store the image and the image feature data to form an update of the patrol image gallery; forming an image duplicate checking service to perform duplicate checking and comparison on the input image in the patrol image gallery according to a feature matching strategy. By establishing an index basis for the image source data based on the extracted image feature differences and forming a patrol image gallery, the duplicate checking process of incremental images effectively combines the retrieval process and the image features, so that when batch duplicate checking is performed on incremental images, the parallel retrieval performance and duplicate checking performance can be efficiently maintained, and the duplicate checking performance required for the continuous increase of patrol images in the patrol image gallery can be fully satisfied.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly relates to a patrol image filtering method and a filtering device. Background Art

[0002] In the prior art, in order to ensure the gas use safety of gas users, relevant units stipulate that staff need to regularly conduct safety inspections on users' gas equipment (pipelines, flow meters, valves, etc.) and gas use conditions. During the safety inspection process, patrol personnel need to use gas-specific mobile devices (PDAs) to take pictures and retain them at key positions. Due to the huge number of end-users and the scale of the pipe network, the number of newly added patrol images per month exceeds one million. To ensure that the images uploaded by patrol personnel comply with the safety inspection regulations and are true and valid, the quality inspection team needs to conduct 100% coverage image quality inspection work to avoid fraud such as duplicate upload of photos and upload of the same photo after cropping. For millions of incremental images, it is impossible to achieve 100% full coverage of quality inspection through manual comparison one by one under the limited time and manpower. Summary of the Invention

[0003] In view of the above problems, embodiments of the present invention provide a patrol image filtering method and a filtering device, which solve the technical problems of low efficiency of existing image quality inspection and incomplete coverage of quality inspection.

[0004] The patrol image filtering method of the embodiment of the present invention includes:

[0005] Form a feature extraction service, extract features and quantify the image to form image feature data;

[0006] Form a gallery storage service, establish a storage structure and an index structure to store the image and the image feature data respectively to form an update of the patrol image gallery;

[0007] Form an image duplicate check service, and perform duplicate check and comparison on the input image in the patrol image gallery according to the feature matching strategy.

[0008] In an embodiment of the present invention, the feature extraction service includes:

[0009] Receive a patrol image, collect the timing information and geographical information in the patrol image to form on-site identification data for collecting the patrol image;

[0010] Extract the global feature descriptor and the local feature descriptor of the patrol image through the DOLG model;

[0011] Use the feature descriptor to form image feature data.

[0012] In an embodiment of the present invention, the gallery storage service includes:

[0013] Perform hashing on the incoming inspection images to form a unique identifier;

[0014] Perform base64 encoding on the incoming inspection images to form image string data;

[0015] Store the image feature data in an index structure to form retrieval data, and store the image string data in a storage structure to form inspection data.

[0016] In one embodiment of the present invention, the image duplicate checking service includes:

[0017] Calculate the Hamming distance one by one between the image feature data of the input image and the retrieval data of the inspection image library;

[0018] According to the set first Hamming distance threshold range, determine the first alternative inspection image set similar to the image features of the input image;

[0019] Extract the existing inspection images from the first alternative inspection image set to form a recommended comparison image set for manual verification.

[0020] In one embodiment of the present invention, the image duplicate checking service includes:

[0021] Extract the global feature data in the image feature data of the input image and calculate the Hamming distance one by one with the global feature data in the retrieval data of the inspection image library;

[0022] According to the set second Hamming distance threshold range, determine the second alternative inspection image set similar to the global image features of the input image;

[0023] Extract the local feature data in the image feature data of the input image and calculate the Hamming distance one by one with the local feature data in the retrieval data of the second alternative inspection image set;

[0024] Sort according to the Hamming distance in the second alternative inspection image set, and according to the set third Hamming distance threshold range, determine the third alternative inspection image set similar to the local image features of the input image;

[0025] Extract the existing inspection images from the third alternative inspection image set to form a recommended comparison image set for manual verification.

[0026] In one embodiment of the present invention, the image duplicate checking service includes:

[0027] Extract the global feature data in the image feature data of the input image and calculate the Hamming distance one by one with the global feature data in the retrieval data of the inspection image library;

[0028] According to the set second Hamming distance threshold range, determine the second alternative inspection image set similar to the global image features of the input image;

[0029] Calculate the Hamming distance one by one between the local feature data in the image feature data of the input image and the local feature data in the retrieval data of the inspection gallery.

[0030] Determine the fourth set of alternative inspection images similar to the local image features of the input image according to the set fourth Hamming distance threshold range.

[0031] Sort the overlapping alternative inspection images in the second set of alternative inspection images and the fourth set of alternative inspection images to form the fifth set of alternative inspection images.

[0032] Extract the existing inspection images according to the alternative inspection images with the prior selected sorting in the fifth set of alternative inspection images to form a recommended comparison image set for manual verification.

[0033] The inspection image filtering device according to the embodiment of the present invention includes:

[0034] A feature extraction module, configured to form a feature extraction service, extract and quantify features of an image to form image feature data.

[0035] A gallery update module, configured to form a gallery storage service, establish a storage structure and an index structure to store images and image feature data respectively to form an update of the inspection gallery.

[0036] A feature duplicate checking module, configured to form an image duplicate checking service, and perform duplicate checking and comparison on the input image in the inspection gallery according to a feature matching strategy.

[0037] The inspection image filtering method according to the embodiment of the present invention includes:

[0038] Receive a new inspection image, and use the feature extraction service to form corresponding new image feature data.

[0039] Use the image duplicate checking service to traverse and compare the new image feature data in the inspection gallery to determine a set of alternative inspection images, and select and submit for manual verification and determination according to the number of recommended comparison images.

[0040] Use the gallery storage service to store the new inspection images that pass the determination in an orderly manner for images and image feature data, and form an incremental update of the inspection gallery.

[0041] The inspection image filtering device according to the embodiment of the present invention includes:

[0042] A new image receiving module, configured to receive a new inspection image, and use the feature extraction service to form corresponding new image feature data.

[0043] A new image duplicate checking module is added, which is used to traverse and compare the newly added image feature data in the inspection image library by using the image duplicate checking service to determine an alternative inspection image set, and select and submit it for manual verification and determination according to the recommended number of comparison images;

[0044] A new image storage module in the library is added, which is used to store the determined newly added inspection images and their image feature data in an orderly manner by using the library storage service, forming an incremental update of the inspection image library.

[0045] The inspection image filtering device according to the embodiment of the present invention includes:

[0046] A memory, which is used to store the corresponding program code during the processing of the above-mentioned inspection image filtering method;

[0047] A processor, which is used to execute the program code.

[0048] The inspection image filtering method and filtering device according to the embodiment of the present invention establish an index basis for the image source data by extracting the image feature differences and form an inspection image library, realizing the effective combination of the retrieval process and image features in the duplicate checking process of incremental images. When batch checking the duplicate of incremental images, the parallel retrieval performance and duplicate checking performance can be efficiently maintained, and the duplicate checking performance required for the continuous increase of inspection images in the inspection image library can be fully satisfied. Description of the Drawings

[0049] Figure 1 The figure shows a schematic flowchart of an inspection image filtering method according to an embodiment of the present invention.

[0050] Figure 2 The figure shows a schematic architecture diagram of an inspection image filtering device according to an embodiment of the present invention.

[0051] Figure 3 The figure shows a schematic flowchart of an inspection image filtering method according to an embodiment of the present invention.

[0052] Figure 4 The figure shows an application schematic diagram of an inspection image filtering method according to an embodiment of the present invention.

[0053] Figure 5 The figure shows a schematic flowchart of an inspection image filtering device according to an embodiment of the present invention. Detailed Embodiments

[0054] To make the purpose, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0055] As shown in the following, a method for filtering inspection images according to an embodiment of the present invention is as follows. Figure 1 In this method, the embodiments of the present invention include: Figure 1 Step 100: Form a feature extraction service to extract and quantify features of the image to form image feature data.

[0056] By using computer image processing technology, the main image features of a digital image can be extracted, such as color distribution, texture orientation, or shape generalization. Through targeted optimization of feature extraction for the image acquisition environment, a good information expression of the on-site content in the image can be obtained. By orderly quantifying the extracted features to form feature data, an efficient and accurate data basis can be provided for comparison between images.

[0057] Step 200: Form a gallery storage service to establish a storage structure and an index structure to store the image and the image feature data respectively, forming an update of the inspection image gallery.

[0058] The images include historical existing images and incremental images of inspections. Through the storage structure, an orderly storage of the image source data is formed. Through the index structure, the image feature data is converted into effective index data of the image source data. The storage structure and the index structure of the inspection image gallery can be formed by using general database technology or the storage and retrieval characteristics of a specific general data structure. The image source data can be different bearing forms of on-site collected image information.

[0059] Step 300: Form an image duplicate check service to perform duplicate check comparison on the input image in the inspection image gallery according to a feature matching strategy.

[0060] The feature matching strategy includes, but is not limited to, feature comparison type limitation, feature comparison threshold setting, feature comparison logic setting, etc. Through the feature matching strategy, the adjustment of the duplicate check comparison logic and efficiency is formed.

[0061] The method for filtering inspection images according to the embodiments of the present invention establishes an index basis for the image source data through the extracted image feature differences and forms an inspection image gallery, realizing the duplicate check process of incremental images, effectively combining the retrieval process and the image features. When performing batch duplicate checks on incremental images, the parallel retrieval performance and duplicate check performance can be efficiently maintained, fully meeting the duplicate check performance required for the continuous increase of inspection images in the inspection image gallery.

[0062] As shown in the following, in an embodiment of the present invention, step 100 includes:

[0063] As follows Figure 1 Step 110: Receive an inspection image, and collect the timing information and geographical information in the inspection image to form on-site identification data for collecting the inspection image.

[0064] Step 110: Receive an inspection image, and collect the timing information and geographical information in the inspection image to form on-site identification data for collecting the inspection image.

[0065] The timing information and geographical information are formed by the timestamp information and geographical coordinate information generated when the gas special mobile device executes the photographing instruction during the formation of the inspection image. The on-site identification data can be used as the on-site feature data of the inspection image.

[0066] Step 120: Extract the global feature descriptor and local feature descriptor of the inspection image through the DOLG model.

[0067] The DOLG model is a framework for fusing local and global features based on deep orthogonality. This algorithm first uses multi-scale convolution and self-attention methods to centrally extract representative local information. Then, it extracts the components orthogonal to the global image representation from the local information. Finally, the orthogonal components are complementarily connected with the global representation method and then aggregated to generate the final representation.

[0068] Specifically, 256-dimensional global representation descriptors and 14x14 local representation descriptors with 256 dimensions containing on-site appearance information such as texture and shape are extracted through the DOLG model.

[0069] Step 130: Use the feature descriptors to form image feature data.

[0070] All the feature descriptors are binary quantized and converted into image feature data in the form of a binary fixed-length sequence. The image feature data includes global feature data of the binary fixed-length sequence (the first part of the binary fixed-length sequence) and local feature data of the binary fixed-length sequence (the second part of the binary fixed-length sequence).

[0071] In an embodiment of the present invention, the on-site identification data is formatted as an additional component of the image source data.

[0072] In an embodiment of the present invention, the on-site identification data is converted into a binary fixed-length sequence and used as an additional component of the image feature data.

[0073] The inspection image filtering method in the embodiments of the present invention mainly uses global feature descriptors and local feature descriptors to form a formatted binary fixed-length sequence to represent image features, so that the image features of the inspection image have a general basis for flexible application on the basis of obtaining relatively complete recognition and extraction.

[0074] As Figure 1 shown, in an embodiment of the present invention, step 200 includes:

[0075] Step 210: Perform hashing processing on the warehoused inspection images to form unique identifiers.

[0076] The hashing processing can be performed using a hashing algorithm. The hashing processing can include the on-site identification data of the warehoused inspection images.

[0077] Step 220: Perform base64 encoding on the in-storage inspection image to form image string data.

[0078] Stringifying the in-storage inspection image can simplify the storage structure and avoid introducing complex data types. The on-site identification data and the image string data can form a data connection.

[0079] Step 230: Store the image feature data into an index structure to form retrieval data, and store the image string data into a storage structure to form inspection data.

[0080] An incremental update of the inspection image library is formed by writing the image feature data and the image string data.

[0081] The inspection image filtering method according to the embodiment of the present invention associates image features with retrieval information by constructing retrieval data, forming the technical effect of image feature comparison and simultaneous image filtering. At the same time, the image source data is stored using string data, reducing the design difficulty and operation overhead of the inspection image library.

[0082] As Figure 1 shown, in an embodiment of the present invention, step 300 includes:

[0083] Step 310: Calculate the Hamming distance one by one between the image feature data of the input image and the retrieval data of the inspection image library.

[0084] Taking advantage of the binary fixed-length sequence characteristic of the image feature data, an efficient calculation of the Hamming distance between the image feature data of the input image and the image feature data in the retrieval data can be formed.

[0085] According to the preset data setting rules, the global feature data formed by the global representation descriptors included in the image feature data of the input image, the local feature data formed by the local representation descriptors, and the on-site identification data should correspond to the data types of the image feature data of the corresponding image source data in the inspection image library.

[0086] Step 320: Determine a first set of alternative inspection images similar to the image features of the input image according to the set first Hamming distance threshold range.

[0087] In an embodiment of the present invention, the first Hamming distance threshold range can be the first 5 alternative inspection images with the closest (i.e., most similar) Hamming distance after traversing and comparing with the retrieval data.

[0088] Step 330: Extract the existing inspection images from the first set of alternative inspection images to form a recommended comparison image set for manual verification.

[0089] The inspection image filtering method according to the embodiment of the present invention uses machine automatic data filtering to quickly narrow down the alternative range of similar images, and combines with manual verification to accurately form a similarity judgment using expert experience, which can effectively improve the duplicate checking efficiency.

[0090] As Figure 1 shown, in an embodiment of the present invention, step 300 further includes:

[0091] Step 340: Calculate the Hamming distance one by one between the global feature data in the image feature data of the input image and the global feature data in the retrieval data of the inspection image library.

[0092] Step 345: Determine a second alternative inspection image set similar to the global image features of the input image according to the set second Hamming distance threshold range.

[0093] In an embodiment of the present invention, the second Hamming distance threshold range may be the first 100 alternative inspection images with the closest Hamming distance after traversing and comparing with the global feature data in the retrieval data.

[0094] Step 350: Calculate the Hamming distance one by one between the local feature data in the image feature data of the input image and the local feature data in the retrieval data of the second alternative inspection image set.

[0095] Step 355: Sort according to the Hamming distance in the second alternative inspection image set, and determine a third alternative inspection image set similar to the local image features of the input image according to the set third Hamming distance threshold range.

[0096] In an embodiment of the present invention, the third Hamming distance threshold range may be the first 3 alternative inspection images with the closest Hamming distance after traversing and comparing with the local feature data in the second alternative inspection image set.

[0097] Step 360: Extract existing inspection images according to the third alternative inspection image set to form a recommended comparison image set for manual verification.

[0098] The inspection image filtering method according to the embodiment of the present invention uses secondary progressive machine automatic data filtering to quickly narrow down the alternative range of similar images, and combines with manual verification to accurately form a similarity judgment using expert experience, which can effectively improve the duplicate checking efficiency.

[0099] As Figure 1 shown, in an embodiment of the present invention, step 300 further includes:

[0100] Step 370: Calculate the Hamming distance one by one between the global feature data in the image feature data of the input image and the global feature data in the retrieval data of the inspection gallery.

[0101] Step 375: Determine a second set of alternative inspection images similar to the global image features of the input image according to the set second Hamming distance threshold range.

[0102] In an embodiment of the present invention, the second Hamming distance threshold range may be the top 10 alternative inspection images with the closest Hamming distance after traversing and comparing with the global feature data in the retrieval data.

[0103] Step 380: Calculate the Hamming distance one by one between the local feature data in the image feature data of the input image and the local feature data in the retrieval data of the inspection gallery.

[0104] Step 385: Determine a fourth set of alternative inspection images similar to the local image features of the input image according to the set fourth Hamming distance threshold range.

[0105] In an embodiment of the present invention, the fourth Hamming distance threshold range may be the top 10 alternative inspection images with the closest Hamming distance after traversing and comparing with the local feature data in the retrieval data.

[0106] Step 390: Sort the overlapping alternative inspection images in the second set of alternative inspection images and the fourth set of alternative inspection images to form a fifth set of alternative inspection images.

[0107] The overlapping alternative inspection images are sorted according to the sorting weights of the second set of alternative inspection images and the fourth set of alternative inspection images. For example, the sorting weights of the second set of alternative inspection images and the fourth set of alternative inspection images are 0.3 and 0.6. If an overlapping alternative inspection image is ranked 8 and 2 in the two sets, then its ranking in the fifth set of alternative inspection images is 8 * 0.3 + 2 * 0.6 = 3.6 ≈ 4.

[0108] Step 395: Extract the existing inspection images according to the alternative inspection images with the selected top rankings in the fifth set of alternative inspection images to form a recommended comparison image set for manual verification.

[0109] The inspection image filtering method in the embodiment of the present invention uses parallel traversal retrieval of differential image features for machine automatic data filtering to quickly narrow down the alternative range of similar images. By combining fine sorting and manual verification and using expert experience to accurately form similarity judgments, the duplicate checking efficiency can be effectively improved.

[0110] In an embodiment of the present invention, the comparison accuracy of forming a recommended comparison image from existing inspection images is improved through the combination of the above three duplicate check comparison processes.

[0111] In practical applications, the above image duplicate check service can accurately compare and confirm newly added inspection images with consistent content. When the recommended comparison image is unique, manual verification is not required.

[0112] An inspection image filtering device according to an embodiment of the present invention is as Figure 2 shown. In Figure 2 , an embodiment of the present invention includes:

[0113] A feature extraction module 10, configured to form a feature extraction service, extract and quantify features of an image to form image feature data;

[0114] A library update module 20, configured to form a library storage service, establish a storage structure and an index structure to respectively store images and image feature data to form an updated inspection library;

[0115] A feature duplicate check module 30, configured to form an image duplicate check service, and perform duplicate check comparison on the input image in the inspection library according to a feature matching strategy.

[0116] As Figure 2 shown, in an embodiment of the present invention, the feature extraction module 10 includes:

[0117] A field feature extraction unit 11, configured to receive an inspection image, and collect temporal information and geographical information in the inspection image to form on-site identification data of the collected inspection image;

[0118] An image feature extraction unit 12, configured to extract a global feature descriptor and a local feature descriptor of the inspection image through a DOLG model;

[0119] An image feature encoding unit 13, configured to form image feature data by using the feature descriptor.

[0120] As Figure 2 shown, in an embodiment of the present invention, the library update module 20 includes:

[0121] An image identification unit 21, configured to perform hashing processing on the incoming inspection image to form a unique identifier;

[0122] An image conversion unit 22, configured to perform base64 encoding on the incoming inspection image to form image string data;

[0123] A data mapping unit 23, configured to store the image feature data into an index structure to form retrieval data, and store the image string data into a storage structure to form inspection data.

[0124] AsFigure 2 As shown in the figure, in an embodiment of the present invention, the feature duplicate checking module 30 includes:

[0125] A first comparison unit 31, configured to calculate the Hamming distance one by one according to the image feature data of the input image and the retrieval data of the inspection image library;

[0126] A first filtering unit 32, configured to determine a first alternative inspection image set similar to the image features of the input image according to a set first Hamming distance threshold range;

[0127] A first recommendation unit 33, configured to extract existing inspection images from the first alternative inspection image set to form a recommended comparison image set for manual verification.

[0128] As Figure 2 shown in the figure, in an embodiment of the present invention, the feature duplicate checking module 30 further includes:

[0129] A second comparison unit 34a, configured to calculate the Hamming distance one by one for the global feature data in the image feature data of the input image and the global feature data in the retrieval data of the inspection image library;

[0130] A second filtering unit 34b, configured to determine a second alternative inspection image set similar to the global image features of the input image according to a set second Hamming distance threshold range;

[0131] A progressive comparison unit 34c, configured to calculate the Hamming distance one by one for the local feature data in the image feature data of the input image and the local feature data in the retrieval data of the second alternative inspection image set;

[0132] A progressive filtering unit 34d, configured to sort according to the Hamming distance in the second alternative inspection image set, and determine a third alternative inspection image set similar to the local image features of the input image according to a set third Hamming distance threshold range;

[0133] A second recommendation unit 34e, configured to extract existing inspection images from the third alternative inspection image set to form a recommended comparison image set for manual verification.

[0134] As Figure 2 shown in the figure, in an embodiment of the present invention, the feature duplicate checking module 30 further includes:

[0135] A third comparison unit 37a, configured to calculate the Hamming distance one by one for the global feature data in the image feature data of the input image and the global feature data in the retrieval data of the inspection image library;

[0136] A third filtering unit 37b, configured to determine a second set of alternative inspection images similar to the global image features of the input image according to a set second Hamming distance threshold range;

[0137] A fourth comparison unit 37c, configured to calculate the Hamming distance one by one between the local feature data in the image feature data of the input image and the local feature data in the retrieval data of the inspection image library;

[0138] A fourth filtering unit 37d, configured to determine a fourth set of alternative inspection images similar to the local image features of the input image according to a set fourth Hamming distance threshold range;

[0139] A parallel filtering unit 37e, configured to sort the overlapping alternative inspection images in the second set of alternative inspection images and the fourth set of alternative inspection images to form a fifth set of alternative inspection images;

[0140] A third recommendation unit 37f, configured to extract existing inspection images from the alternative inspection images sorted first in the fifth set of alternative inspection images to form a recommended comparison image set for manual verification.

[0141] An inspection image filtering method according to an embodiment of the present invention is as Figure 3 shown. In Figure 3 , the embodiment of the present invention uses the service provided by the inspection image filtering method in the above embodiment, including:

[0142] Step 400: Receive a new inspection image, and use the feature extraction service to form corresponding new image feature data.

[0143] Step 500: Use the image duplicate checking service to traverse and compare the new image feature data in the inspection image library to determine a set of alternative inspection images, and select to submit manual verification and determination according to the number of recommended comparison images.

[0144] When there is a unique alternative inspection image, it is determined that there is a duplicate image and the new inspection image fails. When the set of alternative inspection images is empty, it is determined that there is no duplicate image and the new inspection image passes. When the number of alternative inspection images > 1, submit manual verification and determine that the new inspection image passes according to the manual approval.

[0145] Step 600: Use the library storage service to store the determined new inspection image and the image feature data in an orderly manner to form an incremental update of the inspection image library.

[0146] The inspection image filtering method according to the embodiments of the present invention forms a flexible processing process for inspection image filtering and inspection image library update. It can not only be used for duplicate checking of newly added inspection images, but also be applied to the initial construction of the inspection image library and the merging of image libraries. The formed library resources can be further utilized and developed as on-site dynamic resources.

[0147] The application of the inspection image filtering method according to an embodiment of the present invention is as Figure 4 shown. In Figure 4 , the inspection pictures of this period are compared with the content of the historical inspection image library, and highly similar historical pictures are automatically found. After manual determination by quality inspection, the verification result is used to determine whether the current inspection picture is compliant and can pass.

[0148] The inspection image filtering device according to an embodiment of the present invention is as Figure 5 shown. In Figure 5 , the embodiments of the present invention include:

[0149] A new image receiving module 40, configured to receive newly added inspection images and form corresponding new image feature data by using a feature extraction service;

[0150] A new image duplicate checking module 50, configured to traverse and compare the new image feature data in the inspection image library by using an image duplicate checking service to determine an alternative inspection image set, and select and submit for manual verification and determination according to the recommended number of comparison images;

[0151] A new image storage module 60, configured to orderly store the determined newly added inspection images and the image feature data by using an image library storage service to form an incremental update of the inspection image library.

[0152] An inspection image filtering device according to an embodiment of the present invention includes:

[0153] A memory, configured to store program codes corresponding to the processing process of the inspection image filtering method in the above embodiments;

[0154] A processor, configured to execute program codes corresponding to the processing process of the inspection image filtering method in the above embodiments.

[0155] The processor may adopt a DSP (Digital Signal Processor) digital signal processor, an FPGA (Field-Programmable Gate Array) field programmable gate array, an MCU (Microcontroller Unit) system board, an SoC (system on a chip) system board, or a PLC (Programmable Logic Controller) minimum system including I / O.

[0156] As described above, it is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for filtering inspection images, characterized in that, it includes: Form a feature extraction service to extract and quantify features from the image to form image feature data, including: Receive the inspection image, collect the timing information and geographical information in the inspection image to form on-site identification data for collecting the inspection image, and use the on-site identification data as an additional component of the image feature data; Extract the global feature descriptor and local feature descriptor of the inspection image through the DOLG model, and extract 256-dimensional global feature descriptors and 256-dimensional local feature descriptors; Use the feature descriptors to form image feature data, and convert the global feature descriptors, local feature descriptors and on-site identification data into binary fixed-length sequences; Form a gallery storage service, establish a storage structure and an index structure to store images and image feature data respectively to form an updated inspection image gallery, including: Perform hashing on the incoming inspection image to form a unique identifier; Perform base64 encoding on the incoming inspection image to form image string data; Store the image feature data in the index structure to form retrieval data, and store the image string data in the storage structure to form inspection data; Form an image duplicate check service, and perform duplicate check comparison on the input image in the inspection image gallery according to the feature matching strategy. The feature matching strategy includes Hamming distance calculation.

2. The inspection image filtering method according to claim 1, characterized in that, the image duplicate check service includes: Perform Hamming distance calculation one by one on the image feature data of the input image and the retrieval data of the inspection image gallery; According to the set first Hamming distance threshold range, determine the first alternative inspection image set similar to the image features of the input image; Extract the existing inspection images according to the first alternative inspection image set to form a recommended comparison image set for manual verification.

3. The inspection image filtering method according to claim 1, characterized in that, the image duplicate check service includes: Extract the global feature data in the image feature data of the input image and perform Hamming distance calculation one by one on the global feature data in the retrieval data of the inspection image gallery; According to the set second Hamming distance threshold range, determine the second alternative inspection image set similar to the global image features of the input image; Extract the local feature data in the image feature data of the input image and perform Hamming distance calculation one by one on the local feature data in the retrieval data of the second alternative inspection image set; Sort according to the Hamming distance in the second alternative inspection image set, and according to the set third Hamming distance threshold range, determine the third alternative inspection image set similar to the local image features of the input image; Extract the existing inspection images according to the third alternative inspection image set to form a recommended comparison image set for manual verification.

4. The inspection image filtering method according to claim 1, characterized in that, the image duplicate check service includes: Extract the global feature data in the image feature data of the input image and perform Hamming distance calculation one by one on the global feature data in the retrieval data of the inspection image gallery; According to the set second Hamming distance threshold range, determine the second alternative inspection image set similar to the global image features of the input image; Calculate the Hamming distance one by one between the local feature data in the image feature data of the input image and the local feature data in the retrieval data of the inspection gallery; Determine the fourth alternative inspection image set similar to the local image features of the input image according to the set fourth Hamming distance threshold range; Sort the overlapping alternative inspection images in the second alternative inspection image set and the fourth alternative inspection image set to form the fifth alternative inspection image set; Extract the existing inspection images according to the alternative inspection images with the earliest selected sorting in the fifth alternative inspection image set to form a recommended comparison image set for manual verification.

5. An inspection image filtering device, Characterized in that, Comprising: A feature extraction module, used to form a feature extraction service, extract and quantify features of an image to form image feature data; including: A field feature extraction unit, used to receive an inspection image, collect the timing information and geographical information in the inspection image to form the on-site identification data of the collected inspection image, and use the on-site identification data as an additional component of the image feature data; An image feature extraction unit, used to extract the global feature descriptor and local feature descriptor of the inspection image through the DOLG model, and extract a 256-dimensional global feature descriptor and a 256-dimensional local feature descriptor; An image feature encoding unit, used to form image feature data by using the feature descriptor, and convert the global feature descriptor, local feature descriptor and on-site identification data into a binary fixed-length sequence; A gallery update module, used to form a gallery storage service, establish a storage structure and an index structure to store images and image feature data respectively to form an update of the inspection gallery, including: An image identification unit, used to perform a hash process on the incoming inspection image to form a unique identifier; An image conversion unit, used to perform base64 encoding on the incoming inspection image to form image string data; A data mapping unit, used to store the image feature data into the index structure to form retrieval data, and store the image string data into the storage structure to form inspection data; A feature duplicate checking module, used to form an image duplicate checking service, perform duplicate checking and comparison on the input image in the inspection gallery according to the feature matching strategy, and the feature matching strategy includes Hamming distance calculation.

6. An inspection image filtering method, using the inspection image filtering method as described in claim 1, Characterized in that, Comprising: Receive a new inspection image, and use the feature extraction service to form corresponding new image feature data; Use the image duplicate checking service to traverse and compare the new image feature data in the inspection gallery to determine the alternative inspection image set, and select and submit for manual verification and determination according to the number of recommended comparison images; Use the gallery storage service to store the new inspection images that pass the determination in an orderly manner for images and image feature data, and form an incremental update of the inspection gallery.

7. An inspection image filtering device, using the inspection image filtering device as described in claim 5, Characterized in that, Comprising: A new image receiving module, used to receive a new inspection image, and use the feature extraction service to form corresponding new image feature data; A new image duplicate checking module is added, which is used to traverse and compare the newly added image feature data in the inspection image library by using the image duplicate checking service to determine an alternative inspection image set, and select and submit it for manual verification and determination according to the recommended number of comparison images; A new image storage module is added, which is used to orderly store the determined newly added inspection images and their image feature data by using the image library storage service, forming an incremental update of the inspection image library.

8. An inspection image filtering device, characterized in that, it includes: a memory for storing the corresponding program code during the processing of the inspection image filtering method according to any one of claims 1 to 4, 6; a processor for executing the program code.

Images