Nondestructive detection image retrieval method and detection image processing system

Abstract

The invention discloses a nondestructive detection image retrieval method and a detection image processing system. The method comprises the following steps: performing remote connection with scanning detection equipment; monitoring the detection image generated by the scanning detection equipment, and storing the detection image meeting the storage condition through the remote connection; and in response to at least one to-be-retrieved picture uploaded by a user, obtaining a similar picture from the stored detection images based on the to-be-retrieved picture, the similar picture being a retrieval image similar to the to-be-retrieved picture. According to the technology, the server is remotely connected with the scanning detection equipment, and the detection image is monitored and stored. According to the method, the requirement of processing scanning detection equipment of different models at the same time can be met, and the detection images generated by the scanning detection equipment of different models can be stored in the same database. Therefore, the storage is simpler and more convenient, and the required detection image can be conveniently retrieved at any time.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a non-destructive testing image retrieval method and a testing image processing system. Background Art

[0002] With the development of industrial nondestructive testing, more and more types of testing equipment have sprung up. Due to the different storage paths of different brands and models of testing equipment, it is difficult to collect and summarize data.

[0003] Accordingly, the art needs a new detection image processing solution to solve the above problems. Summary of the invention

[0004] In order to overcome the above-mentioned defects, the present invention is proposed to provide a solution or at least partially solve the multi-model equipment in the prior art.

[0005] In a first aspect, the present invention provides a method for retrieving nondestructive testing images, the method comprising: remotely connecting to a scanning testing device; monitoring the testing images generated by the scanning testing device, and storing the testing images that meet the storage conditions through the remote connection; in response to at least one image to be retrieved uploaded by a user, obtaining a similar image from the stored testing images based on the image to be retrieved, wherein the similar image is a retrieval image similar to the image to be retrieved.

[0006] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the obtaining of a similar image from a stored detection image based on the image to be retrieved includes: reducing the image to be retrieved based on a preset resolution to obtain the grayscale value and average grayscale value of each pixel in the reduced image to be retrieved; based on the grayscale value of each pixel, comparing with the average grayscale value respectively to obtain a first feature sequence, and obtaining a second feature sequence corresponding to the stored detection image, the first feature sequence is used to characterize the comparison result between the grayscale value of each pixel and the average grayscale value, and the second feature sequence is based on the stored detection image. The stored detection image is reduced and is used to characterize the comparison result between the grayscale value of each pixel in the reduced stored detection image and the average grayscale value, and the reduced image to be retrieved and the reduced stored detection image have the same resolution; based on the first feature sequence, the second feature sequence corresponding to the detection image is respectively evaluated for similarity to obtain similarity values, wherein one of the detection images corresponds to one similarity value; based on the similarity value, K detection images are determined from the stored detection images, and the K detection images are the detection images corresponding to the first K similarity values ​​after the similarity values ​​are sorted from large to small.

[0007] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, based on the first feature sequence, a similarity evaluation is performed on the second feature sequence of the corresponding detection image to obtain a similarity value, including: comparing the symbols at corresponding positions of the first feature sequence and the second feature sequence, and counting the number of identical symbols between the two; and taking the number of identical symbols as the similarity.

[0008] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the image to be retrieved is reduced based on a preset resolution to obtain the grayscale value of each pixel in the reduced image to be retrieved and the average grayscale value, including: reducing the image to be retrieved based on a preset resolution to obtain the grayscale value of each pixel after the reduction; averaging the grayscale values ​​of the pixels to obtain the average grayscale value.

[0009] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, detection images that meet the storage conditions are stored through the remote connection, including: renaming the detection images at least according to the order that meets the storage conditions and the original names of the detection images, and performing image optimization processing and compression on the detection images; storing the renamed and pre-processed detection images.

[0010] As an alternative or supplement to the above scheme, in a method according to one embodiment of the present invention, the image optimization processing includes: filtering the detection image using a composite image filtering method and processing the detection image using an image refinement technology, and the compression includes: reducing the color depth of the detection image, wherein the composite image filtering method includes at least two filtering methods including anisotropic filtering, multi-scale enhancement or wavelet packet filtering, and the image refinement technology includes at least one of: adaptive local histogram, intensity coefficient optimization, smoothing processing, edge clipping, and threshold value optimization.

[0011] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the detection images generated by the scanning detection device are monitored, and the detection images that meet the storage conditions are stored through the remote connection, including: whenever a new detection image is generated in the monitored area, it is determined whether the newly generated detection image meets the preset volume replication conditions; if the newly generated detection image meets the preset volume replication conditions, the corresponding detection image is copied to the database.

[0012] As an alternative or supplement to the above scheme, in a method according to one embodiment of the present invention, the monitoring of the detection image generated by the scanning detection device includes: acquiring the format of the detection image and the area to be monitored; and monitoring new files that meet the format of the detection image in the area to be monitored in the scanning detection device.

[0013] In a second aspect, a detection image processing system is provided, the system comprising: a remote connection and storage module: remotely connected to a scanning detection device; monitoring the detection images generated by the scanning detection device, and storing the detection images that meet the storage conditions through the remote connection; a retrieval module: in response to at least one image to be retrieved uploaded by a user, obtaining similar images from the stored detection images based on the image to be retrieved, wherein the similar images are images similar to the image to be retrieved. A defect recognition module: using a trained defect recognition neural network to perform defect recognition on the detection image to obtain a defect recognition result.

[0014] As an alternative or supplement to the above scheme, in a method according to an embodiment of the present invention, the use of a trained defect recognition neural network to perform defect recognition on a detection image to obtain a defect recognition result includes: using a trained defect recognition neural network to perform defect recognition on a detection image to obtain a defect recognition frame list, wherein the defect recognition frame list includes M defect recognition frames arranged in descending order of confidence, and the defect recognition frames in the defect recognition frame list all correspond to the same defect category; selecting the defect recognition frame with the highest confidence from the defect recognition frame list for marking, and calculating the confidence ratio between the defect recognition frame with the highest confidence and the remaining defect recognition frames in the defect recognition frame list; The defect recognition frames with an intersection-and-union ratio greater than a preset threshold and the defect recognition frames with the highest confidence are removed from the defect recognition frame list; the defect recognition frames with the highest confidence are selected from the removed defect recognition frame list for marking, and the intersection-and-union ratio of the defect recognition frame with the highest confidence and the remaining defect recognition frames in the defect recognition frame list is calculated, and the defect recognition frames with an intersection-and-union ratio greater than a preset threshold and the defect recognition frames with the highest confidence are removed from the defect recognition frame list; this step is repeated until there are no defect recognition frames in the defect recognition frame list; the marked defect recognition frames are summarized and output as defect recognition results together with the defect categories and confidences corresponding to the defect recognition frames.

[0015] In a third aspect, a control device is provided, which includes a processor and a storage device, wherein the storage device is suitable for storing multiple computer programs, and the computer programs are suitable for being loaded and run by the processor to execute the non-destructive testing image retrieval method described in any one of the technical solutions of the above-mentioned non-destructive testing image retrieval method.

[0016] In a fourth aspect, a computer-readable storage medium is provided, wherein a plurality of computer programs are stored in the computer-readable storage medium, wherein the computer programs are suitable for being loaded and run by a processor to execute the non-destructive testing image retrieval method described in any one of the technical solutions of the above-mentioned non-destructive testing image retrieval method.

[0017] The above one or more technical solutions of the present invention have at least one or more of the following beneficial effects:

[0018] In the technical solution of the present invention, by remotely connecting the server with the scanning detection device and monitoring and storing the detection images, the requirements of processing scanning detection devices of different models at the same time can be met, and the detection images generated by scanning detection devices of different models can be stored in the same database, making the storage simpler and more convenient, and convenient to retrieve the required detection images at any time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0021] Figure 1 is a schematic flow chart of main steps of a nondestructive testing image retrieval method according to an embodiment of the present invention;

[0022] Figure 2 It is a schematic diagram of processing the detection image;

[0023] Figure 3 It is a timing diagram for processing the detection image;

[0024] Figure 4 It is a schematic diagram of the processing flow of the newly generated detection image before copying it to the database;

[0025] Figure 5 It is a schematic diagram of the steps for preprocessing the detection image;

[0026] Figure 6 It is an interactive schematic diagram of obtaining similar images from the image to be retrieved;

[0027] Figure 7 It is a flowchart of the retrieval method;

[0028] Figure 8 It is a schematic diagram of obtaining a feature sequence from a stored detection image;

[0029] Fig. 9 is a main structural block diagram of a detection image processing system according to an embodiment of the present invention;

[0030] List of reference numerals:

[0031] 11: Remote connection and storage module; 12: Retrieval module; 13: Defect identification module. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] Here, some terms related to the present invention are explained first.

[0035] Industrial nondestructive imaging testing: Industrial nondestructive imaging testing (NDT) is a technology used to detect internal defects of industrial products or materials. It uses imaging technologies such as X-rays, ultrasound, and infrared to detect objects, and then images the inside of the detected object to achieve the purpose of not damaging the object being tested. This method can reveal defects such as cracks, pores, and foreign matter inside the product, ensuring product quality and safety, and can produce internal images that are easy to observe. It is widely used in manufacturing, aerospace, energy and other fields.

[0036] At present, in the traditional process of non-destructive image testing of composite materials, scanning testing equipment is used to scan the composite materials. Each scan will obtain a corresponding test image, which will be copied to a CD or uploaded to the cloud.

[0037] However, scanning and testing equipment of different brands or models have their own storage paths and unique file naming habits. If a company uses multiple different brands and models of scanning and testing equipment at the same time, the storage paths and names will be diverse, which is not conducive to the unified analysis and management of data. For example, Company A is a large shipbuilding company. Before the various components of a ship are assembled, each component must be inspected. Company A has a team responsible for inspecting each component, and each inspection team is responsible for inspecting one component. However, each different team is limited by the characteristics of the inspected components and uses scanning and testing equipment of different brands and models. When scanning each component, the scanning and testing equipment used by each team will generate multiple inspection images, and each model of scanning and testing equipment will process and store the inspection images according to its own rules. Then it is very difficult for Company A to summarize and analyze the images generated by these different models of scanning and testing equipment.

[0038] See attached Figure 1 , Figure 1 FIG. 1 is a flow chart showing the main steps of a method for retrieving nondestructive testing images according to an embodiment of the present invention. Figure 1 As shown, the nondestructive testing image retrieval method in the embodiment of the present invention mainly includes the following steps S10 to S30.

[0039] The retrieval method of nondestructive testing images can be divided into two parts according to the scenario. The first part is the process of processing and storing the test images when the scanning test equipment generates the test images. The second part is the specific retrieval process when performing the retrieval.

[0040] The content of the first part corresponds to steps S10-S20. The content of the second part corresponds to step S30. Each step will be described later.

[0041] First of all, let's explain the content of the first part. In order to solve the problems mentioned above, a storage method for detection images is proposed. Figure 2 , Figure 2 This is a schematic diagram of processing a detection image. The server is connected to each scanning detection device by using a remote connection method, and then a monitor is set in each scanning detection device, and the detection image is monitored by the set monitor, and the detection image is processed and stored. The content of the first part includes steps S10-S20, which are as follows:

[0042] Step S10: remotely connect to the scanning and detecting device.

[0043] In response to the user's remote connection operation, the server is remotely connected to the scanning and detection device. The purpose of the remote connection is to send the detection image generated by the scanning and detection device to the database for storage through online transmission when the scanning and detection device is working. The database can be a storage system composed of any storage solution.

[0044] In the process of remotely connecting with the scanning and detection device, first, the server needs to obtain the key connection information of the scanning and detection device, and then remotely connect the server and the scanning and detection device based on the key connection information. In this process, the initiator of the remote connection can be either the server or the scanning and detection device. This is not limited here.

[0045] In one scenario, a user initiates a remote connection request to a scanning and testing device on a server. In this scenario, the user knows the key connection information of the scanning and testing device to be connected in advance. The key connection information is then input into the server, whereupon the server obtains the key connection information and initiates a remote connection request to the scanning and testing device based on the key connection information. When the scanning and testing device receives the remote connection request from the server, it establishes a remote connection with the server.

[0046] In another scenario, the user initiates a remote connection request to the server on the scanning and testing device. In this scenario, the user knows some information about the server, such as the IP address of the server. Afterwards, the IP address of the server is input into the scanning and testing device, so that the scanning and testing device sends a remote connection request to the server, wherein the remote connection request includes key connection information of the scanning and testing device. When the server receives the remote connection request from the scanning and testing device, it establishes a remote connection with the scanning and testing device.

[0047] Among them, the key connection information is the information necessary for the server to establish remote monitoring of the scanning detection device. In some embodiments, the key connection information includes at least connection information and may also include monitoring information. Next, the connection information and monitoring information are described in detail, as follows:

[0048] 1) Connection information

[0049] The connection information is used to complete the connection between the server and the scanning and detecting device. The connection information may include one or more information such as an IP address, authentication information, or a scanning and detecting device name.

[0050] Among them, the IP address can be a public IP or an IP address in the local area network. The server usually locates and connects to the target device (scanning and detection device) through the IP address. Authentication information is first used to provide authentication authority information, so that the server has sufficient authority to connect to the scanning and detection device during the remote connection process. Authentication information is secondly to verify the identity of the user to ensure that only authorized personnel can access and operate the resources of the target device, thereby preventing unauthorized users from invading the system and obtaining sensitive data. Commonly used authentication information includes: user name and password, SSH key or API token, etc. The name of the scanning and detection device is helpful to identify the scanning and detection device to be connected when multiple devices in the local area network have dynamic IP addresses.

[0051] 2) Monitoring information

[0052] The monitoring information is used to provide a basis for the subsequent work of the monitor. The monitoring information can include the file format to be monitored and the area to be monitored. The connection between the server and the scanning and detection device is mainly to allow the server to monitor the scanning and detection device, and the monitoring information provides the file format to be monitored and the area to be monitored. The area to be monitored can tell the monitor where to monitor later, and the file format to be monitored further refines the monitoring scope to ensure that only files of a specific format are monitored.

[0053] In the subsequent monitoring process, the monitor will work based on the monitoring information, and the monitor will monitor the file format to be monitored in the monitored area.

[0054] In this embodiment, the format of the file to be monitored is the format of the detection image generated by the scanning detection device. That is, when the format of the detection image is a Dicom file, the format of the file to be monitored is also a Dicom file.

[0055] The area to be monitored is used to instruct the monitor to monitor a specific location. As can be seen from the previous description, different brands and models of scanning and detection devices will store the detection images in different folders, so the specific setting of the area to be monitored is also related to the scanning and detection device. The area to be monitored can be the bottom-level folder used by the current scanning and detection device to store the detection image, or it can be the upper-level folder of the folder. For example, the scanning and detection device stores the detection image in the folder DailyScans, where the absolute path of DailyScans (from the root directory to the specified folder) is as follows: C:\Scans\DailyScans. The upper-level folders of this folder are C:\Scans and C:. Therefore, the area to be monitored can be any one of C:\Scans\DailyScans, C:\Scans or C:. When the user only determines that the detection image is stored in C:\Scans, but is not sure which specific folder in C:\Scans it is, the area to be monitored can be set to C:\Scans. At this time, the monitor will monitor all other folders stored in C:\Scans. The role of the area to be monitored is to narrow the scope of monitoring, thereby making the monitoring speed faster.

[0056] After the server obtains the above-mentioned key connection information, it can be remotely connected to the scanning device. Specific commands can be used to connect remotely. For example, in the Windows operating system, when establishing a remote connection, the "netuse" command can be used to connect the server to the scanning detection device. The "netuse" command is a network sharing command in Windows, which can establish a network connection between the local computer and the shared resources on the remote computer, so that the user can access the shared folders on the remote computer like accessing local resources. In addition, in Unix / Linux systems, the server and the scanning detection device can be connected via the SSH protocol. In addition, the server and the scanning detection device can also be connected via the Remote Desktop Protocol (RDP), SMB Protocol (ServerMessage Block), VPN (Virtual Private Network) connection and other methods. Those skilled in the art can also select other remote connection methods based on the examples in the above, which are not limited here.

[0057] In summary, after the server receives the remote connection request, it can connect to the scanning detection device based on the remote connection request. After completing the connection with the scanning detection device, a monitor is created based on the monitoring information, and the detection image is stored according to the content monitored by the monitor, as follows:

[0058] Step S20: monitoring the detection images generated by the scanning detection device, and storing the detection images that meet the storage conditions through the remote connection.

[0059] refer to Figure 3 As shown, Figure 3 It is a sequential diagram that the detection image is processed. In the present embodiment, a monitor (FileWatcher) is created to monitor the detection image. The monitor can monitor the area to be monitored mentioned above. The newly generated file is identified. Once the newly generated file meets the format to be monitored (the format of the detection image), then it is determined that the newly generated file is the detection image at this time. Afterwards the newly generated detection image is processed, and the processed detection image is copied into the database.

[0060] Furthermore, in some other embodiments, the monitor monitors the monitored area. Whenever a new detection image is generated in the monitored area, it is also determined whether the newly generated detection image meets the preset volume replication condition. If the newly generated detection image meets the preset volume replication condition, the corresponding detection image is copied to the database.

[0061] like Figure 4 shown. Figure 4 This is a schematic diagram of the processing flow of the newly generated detection image before copying it to the database. When a new detection image is generated in the monitored area, the volume of the detection image is obtained. Then the volume of the detection image is compared with the preset volume threshold. Only when it is larger than the preset volume threshold can subsequent operations be performed.

[0062] Taking the detection image type as a Dicom file as an example, when the monitor finds a newly generated Dicom file in the monitored area, it is determined whether the volume of the Dicom file is greater than a preset volume threshold. For example, whether the volume of the Dicom file is greater than 500KB. If the volume of the Dicom file is greater than 500KB, the Dicom file is copied to the database. If the volume of the Dicom file is less than or equal to 500KB, no copy operation is performed.

[0063] The reason for setting the preset volume replication condition is that the detection image usually has some accompanying files, and these accompanying files are not valid image information, so these accompanying files do not need to be copied. In order to better understand the significance of setting the volume replication condition, an example is given here. In industrial non-destructive image detection, an ultrasonic scanning microscope can be used to obtain a detection image of the material. At this time, the format of the detection image usually uses the Dicom format. The single volume size of this detection image is usually between 20MB and 60MB. The information in the detection image represents the reflection image of the material under different depth information, and this detection image is the file that needs to be added to the database. Some Dicom files with too small a volume are the companion files of the detection image, which only contain metadata (for example, some reports or labels) but no actual image data, so Dicom files that do not meet the volume replication conditions will not be copied to the database. This is why the volume judgment is performed on the detection image.

[0064] In this step, by remotely connecting the server to the scanning and testing device, and monitoring and storing the detection images, the requirements of processing scanning and testing devices of different models at the same time can be met, and the detection images generated by scanning and testing devices of different models can be stored in the same database. The problem of difficult storage of scanning and testing devices of multiple models in the prior art is solved. The storage is made simpler and more convenient, and the required detection images can be retrieved at any time.

[0065] Further, such as Figure 4 As shown. In some embodiments, in the process of storing the detection image in the database, the detection image can also be renamed and preprocessed first, and then the renamed preprocessed detection image is copied to the database. The detection image is renamed to avoid the phenomenon of repeated naming of detection images generated by scanning detection devices of various models, and the detection image is preprocessed to save storage space while retaining rich information. Renaming and preprocessing are described in detail later.

[0066] 1) Rename:

[0067] You can rename the test images by adding a prefix to the file name of the test image. For example, you can add a number in the format of "File000X_" to the file name, where "000X" indicates the order in which the test images are copied to the database, and "_" is a separator used to separate the original file name of the test image. "000X" is a natural number sequence that increases in sequence. For example, the first test image that meets the storage conditions will have "File 0001" added to the original file name, and then the second test image that meets the storage conditions will have "File0002" added to the original file name.

[0068] In addition, the name of the scanning and detection device can be added before the file name to make the prefix information more diverse, thereby avoiding duplication of file names.

[0069] For example, there are two scanning detection devices, namely scanning detection device A and scanning detection device B. Scanning detection device A generates a detection image (hereinafter referred to as detection image A for distinction), and the file name of detection image A is "SAT-I_081_2024-10-17". This file name is divided into three parts by three separators "_", where SAT-I represents ultrasonic detection, "081" is the unique identifier of the detected sample, and "2024-10-17" represents the generation date of the detection image. The scanning detection device also generates a detection image (hereinafter referred to as detection image B for distinction). Coincidentally, the file name of detection image B is also "SAT-I_081_2024-10-17".

[0070] If both test images are copied directly into the database, two test images with the same file name will appear in the database, which will easily cause an error when storing the image or cause one test image to be overwritten by another.

[0071] In this embodiment, two detection images are distinguished by adding prefixes before the original file names, and the more types of prefixes added, the less likely the two detection images are to have repeated nouns. For example, if detection image A is the 2513th detection image copied to the database produced by scanning detection device A, and detection image B is the 1435th detection image copied to the database produced by scanning detection device B. Then the file names of the two detection images after adding the copy order prefix are "File2513_SAT-I_081_2024-10-17" and "File1435_SAT-I_081_2024-10-17", respectively. If the prefix includes not only the copy order, but also the name of the scanning detection device. The file names after adding the copy order and the scanning detection device name prefix are "EQP-A_File2513_SAT-I_081_2024-10-17" and "EQP-B_File1435_SAT-I_081_2024-10-17", respectively. EQP-A indicates that the inspection image is obtained by scanning inspection device A. It can be seen that when the device name is added, even if the original file name is the same and the copy order is the same, the final renamed file name is different.

[0072] It can be seen that by renaming, the features of the detection images transferred to the database can be made more obvious, and the possibility of duplication of detection image names during the storage process is reduced.

[0073] 2) Preprocessing:

[0074] Preprocessing is the further processing of the detection image.

[0075] refer to Figure 5 , Figure 5 Schematic diagram of the steps of image preprocessing. In general, the preprocessing may include first performing image optimization processing on the detection image, and then compressing the detection image after the image optimization processing is completed.

[0076] In some embodiments, before performing other preprocessing methods, the detection image may be first converted into a format that is easy to process subsequently, and then the preprocessing may be performed. For example, when the detection image is in Dicom format, the detection image may be first converted into a Bitmap format that is easy to process, and then the detection image may be processed.

[0077] In order to better understand the preprocessing in this embodiment, the reasons for preprocessing are first explained here. Taking the scanning detection equipment as an ultrasonic scanning microscope as an example, the ultrasonic scanning microscope shoots high-frequency ultrasonic pulses at the sample to be tested, measures the reflected waves to form an image, and thus non-destructively detects the internal structure and defects of the material. However, during use, scanning detection equipment such as ultrasonic scanning microscopes are susceptible to external interference. For example, the ultrasonic signal of an ultrasonic scanning microscope is affected by external noise during propagation. At this time, the detection image may have many noise points, poor contrast, inconsistent illumination in various areas of the image, abnormal exposure of the entire image, and other problems.

[0078] Therefore, the detection image is processed by image optimization processing to remove the noise in the detection image and protect the edges in the detection image, so that the processed detection image is clearer and the image quality is improved. The specific methods of image optimization processing are as follows:

[0079] First, the detection image is processed using the image quality enhancement technology, and then the detection image processed by the image quality enhancement technology is further processed using the image refinement technology.

[0080] Image quality enhancement technology uses filtering methods to filter the detection image. The filtering methods can be: such as anisotropic filtering, multi-scale enhancement or wavelet packet analysis filtering. Anisotropic filtering is mainly used for denoising and retaining edge information. Multi-scale enhancement enhances the details of the detection image by analyzing and processing the features of the detection image at different scales or resolutions. Wavelet packet analysis decomposes the image signal into wavelet packet coefficients of different frequency ranges, so as to process the detection image in more detailed frequency sub-bands for denoising and enhancing the features of the detection image.

[0081] In some embodiments, at least two filtering methods constitute a composite image filtering method, thereby achieving denoising of the detection image and enhancing the display details of the detection image. For example, anisotropic filtering and multi-scale enhancement are used to filter the detection image at the same time, and anisotropic filtering can denoise the detection image, and multi-scale enhancement enhances the image details on the detection image. It is achieved while reducing noise and enhancing details. Similarly, the three filtering methods mentioned above can also be used to filter the detection image at the same time.

[0082] By using image quality enhancement technology, the quality of the inspection image can be improved. The improvement in quality will help improve the accuracy of defect identification.

[0083] Image refinement techniques may include: adaptive local histogram equalization, intensity coefficient optimization, smoothing, edge clipping, threshold optimization, or a combination of one or more of these. Image refinement techniques may be used to process the detected image through a combination of one or more of the above methods, thereby enhancing the image features.

[0084] Among them, adaptive local histogram equalization can detect the contrast of local areas in the image to make the contrast between the defective part and the normal part more obvious. Intensity coefficient optimization can adjust the brightness and contrast of the image pixels in the detected image to make the visual effect of the entire image more balanced. Smoothing is mainly used to remove random noise in the image, reduce the graininess of the image, and make the image look smoother and more natural. Edge cropping can remove unnecessary edge areas in the image to ensure that subsequent processing is concentrated on the effective image part and reduce the interference of background noise or invalid areas. Threshold value optimization divides the image by setting the best threshold and extracts the meaningful part, thereby improving the accuracy of subsequent image analysis during defect diagnosis.

[0085] By processing the image quality and image fineness of the inspection image, the noise of the image is finally suppressed, the details are enhanced, and the contrast of the image is effectively adjusted, and the invalid area in the inspection image is removed. In the subsequent work, the inspection image can more clearly show the possible defects, making the subsequent defect identification easier.

[0086] After the image optimization processing of the inspection image, the optimized inspection image is compressed, as follows: With the development of industrial non-destructive testing technology, the scanning accuracy is also constantly improving. As a result, the fineness of the inspection image produced is higher, which makes the volume of the inspection image larger and larger. If the inspection image is directly stored, it will occupy a lot of storage resources.

[0087] Take the ultrasonic scanning microscope as an example. The detection image output by the ultrasonic scanning microscope is in Dicom format (usually a grayscale image) and is between 20MB and 60MB in size. Storing such a large detection image in a database will take up a lot of storage resources, so a method for compressing an image is proposed, as follows:

[0088] In the process of compressing the detection image, the color depth of the detection image is first reduced, and then the detection image with reduced color depth is stored in a preset format.

[0089] The color depth (bit) is the number of bits used for each color channel. For grayscale images, the color depth refers to the number of gray levels that each pixel can represent. The higher the color depth, the more color details in the image. The inspection image output by the ultrasonic scanning microscope is a 14-bit grayscale image. A 14-bit grayscale image means that each pixel uses 14 bits to represent its grayscale value, which means that each pixel can display 16,384 different gray levels (2 14 =16384). Such a detailed display must occupy a lot of resources. After image optimization, the image features of the inspection image are enhanced, so a reduction in color depth will not affect the judgment of defects. Therefore, the color depth of the inspection image is reduced here. For example, the 14-bit original inspection image can be converted into an 8-bit inspection image. During the conversion process, linear scaling or existing tools (such as open source image processing software ImageJ or by programming to call functions) can be used to reduce the color depth. Generally, the volume of an 8-bit image is 50% smaller than that of a 14-bit image.

[0090] It should be noted that the inspection image compressed into 8bit after image optimization has many advantages over the 8bit inspection image generated directly by scanning inspection equipment. Although it is also an 8bit inspection image, the 8bit inspection image compressed after image optimization is obtained based on the 14bit original inspection image. The original inspection image contains richer information. After image optimization, invalid information and information that interferes with the judgment of defects are eliminated, while valid information is amplified and enhanced. Therefore, the 8bit inspection image compressed after image optimization will retain the valid information in the 14bit original inspection image. The 8bit inspection image directly generated by the scanning inspection equipment may cause information loss due to the small number of gray levels.

[0091] For example, there are two detection images, one of which is a 14-bit detection image that is compressed into an 8-bit detection image after image optimization processing (hereinafter referred to as the 14-bit detection image as detection image C, and the compressed image as detection image D), and the other is an 8-bit detection image generated directly using a scanning detection device (hereinafter referred to as detection image E). If both detection image C and detection image E are overexposed, then since detection image C is a 14-bit image, detection image C contains more information and captures more details and color changes relative to detection image E. When performing image optimization processing, detection image C can more effectively avoid color distortion and noise amplification. Therefore, when adjusting detection image C and detection image E at the same time, the final detection image C will be better than detection image E in terms of details, dynamic range and color restoration. When the image optimization processing is performed, when detection image C is compressed into detection image D, since the detection image C has a better processing effect and obtains more effective details, these effects and details will be retained during compression. As a result, the effect of detection image D is better than that of detection image E.

[0092] In some embodiments, after the color depth is reduced, the detection image is format converted and stored. When storing the detection image, the user can choose whether to convert the detection image into a format. The main function of the format conversion is also to compress the detection image. For example, files in Dicom format are usually used in industrial lossless image detection. In some embodiments, the detection image is converted into one of the formats of PDF or JPEG, and then the detection image after format conversion is stored in a database. Lossy compression will be performed during the conversion of the detection image into PDF and JPEG formats, which will further reduce the volume of the detection image.

[0093] By renaming and preprocessing the test images, the test image file names are made unique to prevent duplicates in the database. And by setting prefixes, it is easier to identify and manage the test images in the database. Preprocessing makes the details of the test images more obvious and rich, making the final test images smaller, thereby significantly reducing the storage space occupied during the storage process.

[0094] like Figure 6 As shown, Figure 6 It is a schematic diagram of the interaction between the image to be retrieved and the similar images. Figure 6In the scenario, the user uses a picture to be retrieved, which is a test image obtained by scanning the test equipment. After the user gets the test image, he cannot easily judge whether there is a defect, so he wants to use the data in the previously collected database as a reference to analyze and judge the test image. Then, the user's demand at this time is to find a test image similar to the test image from the previously stored test images.

[0095] In this embodiment, a fast retrieval method is proposed, so that one or a group of images that are most similar to the image to be retrieved uploaded by the user can be found in the stored detection images. The fast retrieval method is mainly implemented through step S30, which is as follows:

[0096] Step S30: In response to a search operation by the user, similar images are obtained from the stored detection images based on the search operation.

[0097] In this embodiment, the user's search operation includes uploading at least one picture, and the picture uploaded by the user in the search operation is subsequently referred to as the picture to be searched. The picture to be searched here can be a detection image generated by a scanning detection device, or a picture obtained from the Internet or other channels, which is not limited here.

[0098] Similar pictures refer to pictures obtained from the stored detection images that meet the preset similarity conditions with the picture to be retrieved. The user obtains similar pictures through the search operation. In this process, one similar picture or multiple similar pictures may be obtained. Therefore, the number of similar pictures obtained is not limited here.

[0099] Further, such as Figure 7 As shown, Figure 7 is a flowchart of the search method. Specifically, the search method may include steps S301-S305, which are as follows:

[0100] Step S301: reducing the stored detection image based on a preset resolution to obtain the grayscale value of each pixel of the reduced detection image.

[0101] In this embodiment, all the detection images in the database are reduced. Reduction refers to the process of reducing the resolution, that is, the process of reducing the image from more pixels to fewer pixels. Reduction is to simplify the complexity of the subsequent similarity calculation process. The preset resolution is a pre-set resolution value. No matter what the previous resolution of the stored detection image is, it will become the preset resolution size after reduction.

[0102] Assume that a stored detection image has a resolution of 100*100 pixels, with a total of 10,000 pixels. If the preset resolution is 10*10, then the detection image will become a 10*10 image after being reduced. Similarly, the resolution of other stored detection images after being reduced is also 10*10.

[0103] In some embodiments, reducing the resolution of the detection image to a fixed size (reduction) may use methods such as bilinear interpolation, nearest neighbor interpolation, bicubic interpolation, Gaussian pyramid, antialiasing reduction, etc., which are not limited here.

[0104] By reducing the size of the image, the detection becomes very simple. The details of the detected image are reduced, but its overall features (such as the general shape, light and dark areas) are still preserved, which reduces the computing power consumption of subsequent similarity comparisons and reduces the time spent on retrieval.

[0105] Step S302: averaging the grayscale values ​​of each pixel to obtain an average grayscale value.

[0106] The average gray value is a numerical value. Each reduced detection image corresponds to an average gray value. The average gray value is obtained based on the gray value of each pixel and is the average value of the gray value of each pixel.

[0107] When calculating the average gray value, the gray values ​​of each pixel in the reduced detection image are added and averaged to obtain the average gray value.

[0108] like Figure 8 As shown, Figure 8 is a schematic diagram of a stored detection image to obtain a feature sequence. Figure 8 It is divided into 4 parts, namely sub-image a, sub-image b, sub-image c and sub-image d. Sub-image a is the grayscale image of the reduced detection image, and sub-image b is the grayscale value of each pixel in sub-image a.

[0109] Sub-image a is an 8-bit grayscale image, and the resolution of the reduced detection image is 2*2. The grayscale values ​​corresponding to each pixel in sub-image a are 100, 150, 200, and 50. The grayscale values ​​of each pixel are added and averaged: (100+150+200+50) / 4=125. So the final average grayscale value is 125.

[0110] The average grayscale value generated is the brightness baseline of the image. Then the grayscale value of each pixel will be compared with this baseline. Through comparison, the bright and dark areas of the image can be separated, laying the foundation for generating a concise binary representation, thereby further simplifying the image.

[0111] Step S303: obtaining a feature sequence based on the average gray value and the gray value of each pixel.

[0112] In this embodiment, the feature sequence corresponding to the reduced detection image is obtained by comparing the average gray value with the gray value of each pixel.

[0113] like Figure 8 As shown, sub-figure c is a schematic diagram of the comparison between the average grayscale value and the grayscale value. Sub-figure d is a schematic diagram of the result after the comparison, and is also a schematic diagram of the obtained feature sequence. In sub-figure c, the grayscale value of each pixel must be compared with the average grayscale value. In sub-figure d, the final comparison result is shown. When the grayscale value of the pixel is greater than the average grayscale value, it is recorded as a value, such as 1. When the grayscale value of the pixel is less than the average grayscale value, it is recorded as another value, such as 0. Each pixel will get a corresponding value, and when these values ​​are arranged in the order of pixels, they form a feature sequence. The feature sequence can be in matrix form or in a one-dimensional array. In other words, the feature sequence can be represented in the form of a matrix as shown in sub-figure d, or in the form of a one-dimensional array such as [0, 1, 1, 0].

[0114] Since the stored detection image is converted into a short feature sequence, the feature sequence reflects the relative distribution of light and dark in the original image. In the subsequent retrieval process, the light and dark distribution (feature sequence) of the two images can be directly compared to find the image with the closest light and dark distribution, that is, the similar image.

[0115] Furthermore, the obtained feature sequence corresponding to the detection image is stored. After the feature sequence of the detection image is obtained, it is stored in the database. After storage, the feature sequence and the detection image in the database are in a one-to-one correspondence, that is, one feature sequence corresponds to one detection image.

[0116] As described above, during the search process, the user will upload the image to be searched, and then the image to be searched will be matched with the detection image in the database to obtain similar images. When the detection image in the database already has its own feature sequence, it is only necessary to convert the image to be searched into a feature sequence, so that the feature sequence of the image to be searched can be matched with the feature sequence of the detection image, thereby obtaining similarity and similar images. The details are as follows:

[0117] Step S304: Acquire the feature sequence of the image to be retrieved, and perform similarity evaluation on the feature sequence of the image to be retrieved and the feature sequence of the detection image, thereby obtaining a similarity value.

[0118] In this embodiment, the method for obtaining the feature sequence of the image to be retrieved is consistent with the method for obtaining the feature sequence of the detection image. Similarly, the image to be retrieved is reduced according to the preset resolution, and then the grayscale values ​​of each pixel in the reduced image to be retrieved are averaged to obtain the average grayscale value, and then the grayscale value is compared with the average grayscale value to obtain the feature sequence of the image to be retrieved. No further details are given here.

[0119] Since both the image to be retrieved and the detection image are reduced using a preset resolution, the number of pixels of the two images after reduction is the same, and thus the number of symbols in the feature sequence is the same. In some embodiments, the symbols at corresponding positions of the feature sequence of the image to be retrieved and the feature sequence of the detection image are compared to see if they are the same, and the number of identical symbols between the two is counted. The number of identical symbols is used as the similarity.

[0120] For example, the feature sequence of the image to be retrieved is [1011101], and the feature sequence of a detection image is [1001001]. The comparison of these two feature sequences is shown in Table 1.

[0121] Location Feature sequence of the image to be retrieved Detecting feature sequences in images Are they the same? 1 1 1 same 2 0 0 same 3 1 0 different 4 1 0 different 5 1 1 same 6 0 0 same 7 1 1 same

[0122] Table 1

[0123] From Table 1, we can see that there are 5 identical symbols at the same position, so the similarity between the feature sequence of the image to be retrieved and the feature sequence of the detection image is 5. By comparing whether the symbols at corresponding positions are identical, the calculation is simple. When there are a large number of detection images in the database, the similarity between all detection images and the image to be retrieved can be calculated in a shorter time.

[0124] In addition to the above methods, the similarity between two feature sequences can be calculated by calculating Euclidean distance, Manhattan distance (L1), cosine similarity, etc., which are not limited here.

[0125] Step S305: Based on the similarity value, determine K detection images from the stored detection images.

[0126] In this embodiment, the K detection images are detection images corresponding to the first K similarity values ​​after the similarity values ​​are sorted from large to small.

[0127] Users can select the K test images with the highest similarity according to their needs. K is a positive integer. When the K value is selected, K test images will be output. These K test images are similar files.

[0128] It should be pointed out that although the various steps in the above embodiments are described in a specific order, those skilled in the art can understand that in order to achieve the effects of the present invention, different steps do not have to be performed in such an order. They can be performed simultaneously (in parallel) or in other orders. These changes are within the scope of protection of the present invention.

[0129] Furthermore, the present invention also provides a detection image processing system.

[0130] See attached Fig. 9 , Fig. 9 FIG. 1 is a main structural block diagram of a detection image processing system according to an embodiment of the present invention. Fig. 9 As shown, the detection image processing system in the embodiment of the present invention mainly includes a remote connection and storage module 11, a retrieval module 12 and a defect recognition module 13.

[0131] The detection image processing system includes:

[0132] Remote connection and storage module 11: remotely connected to the scanning detection device; monitoring the detection images generated by the scanning detection device, and storing the detection images that meet the storage conditions through the remote connection;

[0133] Retrieval module 12: in response to at least one image to be retrieved uploaded by a user, obtain similar images from the stored detection images based on the image to be retrieved, wherein the similar images are images similar to the image to be retrieved;

[0134] Defect recognition module 13: Use the trained defect recognition neural network to perform defect recognition on the detection image to obtain the defect recognition result.

[0135] The description of the specific functions implemented by the remote connection and storage module 11 and the retrieval module 12 can refer to the description of steps S10-S30.

[0136] The remote connection and storage module 11 and the retrieval module 12 in the above detection image processing system are used to execute Figure 1 The embodiment of the retrieval method for nondestructive testing images shown in the figure has similar technical principles, technical problems solved and technical effects produced. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process and relevant instructions of the remote connection and storage module 11 and the retrieval module 12 can refer to the contents described in the embodiment of the retrieval method for nondestructive testing images, which will not be repeated here.

[0137] The defect recognition module 13 is described in detail here. In the defect recognition module 13, a trained defect recognition neural network is used to perform defect recognition on the detection image to obtain a defect recognition frame list, wherein the defect recognition frame list includes M defect recognition frames arranged in descending order of confidence, and the defect recognition frames in the defect recognition frame list all correspond to the same defect category.

[0138] The defect identification frame with the highest confidence is selected from the defect identification frame list as the output defect identification frame, and the intersection and union ratio of the defect identification frame with the highest confidence and the remaining defect identification frames in the defect identification frame list is calculated, and the defect identification frames whose intersection and union ratio is greater than the preset threshold and the defect identification frame with the highest confidence are removed from the defect identification frame list.

[0139] The defect recognition frame with the highest confidence is selected from the list of defect recognition frames after elimination as the output defect recognition frame, and the intersection and union ratio of the defect recognition frame with the highest confidence and the remaining defect recognition frames in the defect recognition frame list is calculated, and the defect recognition frames whose intersection and union ratio is greater than a preset threshold and the defect recognition frame with the highest confidence are eliminated from the defect recognition frame list; repeat this step until there are no candidate frames in the defect recognition frame list.

[0140] The marked defect recognition boxes are summarized and output together with the defect categories and confidence levels corresponding to the defect recognition boxes as defect recognition results.

[0141] This method can retain the highest confidence or most obvious results and remove redundant or weak response results, making the final defect identification result more accurate.

[0142] Furthermore, the training method of the defect recognition neural network is as follows:

[0143] Obtain image defect samples, and then select negative samples from the image defect samples. Cut the defect area in the negative sample from the original negative sample, and paste the defect area into the positive sample, thereby expanding the negative sample. The defect area is the area where the defect in the negative sample is located.

[0144] Then, the image defect sample is input into the defect recognition neural network, wherein the defect recognition neural network can use the YOLOv8 neural network, and the defect recognition neural network is trained using the back propagation algorithm.

[0145] It is understood by those skilled in the art that the present invention implements all or part of the processes in the method of the above embodiment, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes a computer program, and the computer program can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program. It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electric carrier signals and telecommunication signals.

[0146] Furthermore, the present invention also provides a control device. In an embodiment of a control device according to the present invention, the control device includes a processor and a storage device, the storage device can be configured to store a program for executing the method for retrieving nondestructive testing images of the above method embodiment, and the processor can be configured to execute the program in the storage device, which includes but is not limited to the program for executing the method for retrieving nondestructive testing images of the above method embodiment. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The control device can be a control device device formed by various electronic devices.

[0147] Furthermore, the present invention also provides a computer-readable storage medium. In a computer-readable storage medium embodiment according to the present invention, the computer-readable storage medium can be configured to store a program for executing the method for retrieving nondestructive testing images of the above-mentioned method embodiment, and the program can be loaded and run by a processor to implement the above-mentioned method for retrieving nondestructive testing images. For ease of explanation, only the parts related to the embodiment of the present invention are shown. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium can be a storage device formed by various electronic devices. Optionally, the computer-readable storage medium in the embodiment of the present invention is a non-temporary computer-readable storage medium.

[0148] Further, it should be understood that since the setting of each module is only for illustrating the functional units of the device of the present invention, the physical devices corresponding to these modules may be the processor itself, or a part of the software in the processor, a part of the hardware, or a part of the combination of software and hardware. Therefore, the number of each module in the figure is only schematic.

[0149] Those skilled in the art will appreciate that the modules in the device can be adaptively split or merged. Such splitting or merging of specific modules will not cause the technical solution to deviate from the principle of the present invention, and therefore, the technical solutions after splitting or merging will fall within the protection scope of the present invention.

[0150] It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of this application, and these improvements and modifications should also be regarded as the scope of protection of this application.

Claims

1. A non-destructive testing image retrieval method, characterized in that: include: Remotely connect to scanning and testing equipment; Monitoring the detection images generated by the scanning detection device, and storing the detection images that meet the storage conditions through the remote connection; In response to at least one image to be retrieved uploaded by a user, a similar image is obtained from stored detection images based on the image to be retrieved, wherein the similar image is a retrieval image similar to the image to be retrieved.

2. The method according to claim 1, characterized in that: The obtaining similar images from the stored detection images based on the image to be retrieved includes: Reducing the image to be retrieved based on a preset resolution to obtain a grayscale value and an average grayscale value of each pixel in the reduced image to be retrieved; Based on the grayscale values ​​of the respective pixels, they are compared with the average grayscale value to obtain a first feature sequence, and a second feature sequence corresponding to the stored detection image is obtained, wherein the first feature sequence is used to characterize the comparison result between the grayscale value of the respective pixels and the average grayscale value, and the second feature sequence is obtained after reduction based on the stored detection image, and is used to characterize the comparison result between the grayscale value of each pixel in the reduced stored detection image and the average grayscale value, and the reduced image to be retrieved and the reduced stored detection image have the same resolution; Based on the first feature sequence, respectively perform similarity evaluation on the second feature sequences corresponding to the detection images to obtain similarity values, wherein one detection image corresponds to one similarity value; Based on the similarity values, K detection images are determined from the stored detection images, where the K detection images are detection images corresponding to the first K similarity values ​​after the similarity values ​​are sorted from large to small.

3. The method according to claim 2, characterized in that The step of performing similarity evaluation on the second feature sequences of the corresponding detection images based on the first feature sequences to obtain similarity values ​​includes: Compare symbols at corresponding positions of the first characteristic sequence and the second characteristic sequence, and count the number of identical symbols between the two; The number of the same symbols is regarded as the similarity.

4. The method according to claim 2 or 3, characterized in that: The step of reducing the image to be retrieved based on a preset resolution to obtain a grayscale value and an average grayscale value of each pixel in the reduced image to be retrieved includes: Reducing the image to be retrieved at a preset resolution to obtain a grayscale value of each pixel in the reduced image to be retrieved; The grayscale values ​​of the pixels are averaged to obtain an average grayscale value.

5. The method according to claim 1, characterized in that Storing the detection images that meet the storage conditions through the remote connection includes: renaming the detection images at least according to the order satisfying the storage condition and the original names of the detection images, and performing image optimization processing and compression on the detection images, wherein the image optimization processing is used to enhance the details of the detection images; The renamed and preprocessed detection images are stored.

6. The method according to claim 5, characterized in that The image optimization processing includes: filtering the detection image using a composite image filtering method and processing the detection image using an image refinement technology, and the compression includes: reducing the color depth of the detection image, wherein the composite image filtering method includes at least two filtering methods including anisotropic filtering, multi-scale enhancement or wavelet packet filtering, and the image refinement technology includes at least one of: adaptive local histogram, intensity coefficient optimization, smoothing processing, edge clipping, and threshold value optimization.

7. The method according to claim 1, characterized in that The monitoring of the detection images generated by the scanning detection device and storing the detection images that meet the storage conditions through the remote connection includes: Whenever a new detection image is generated in the monitored area, it is determined whether the newly generated detection image meets the preset volume replication condition; If the newly generated detection image meets the preset volume replication condition, the corresponding detection image is copied into the database.

8. The method according to claim 1, characterized in that The monitoring of the detection image generated by the scanning detection device includes: Get the format of the detected image and the area to be monitored; New files in the to-be-monitored area of ​​the scanning and detecting device that meet the format of the detection image are monitored.

9. A detection image processing system, characterized in that: include: Remote connection and storage module: remote connection with scanning and detection equipment; Monitoring the detection images generated by the scanning detection device, and storing the detection images that meet the storage conditions through the remote connection; A retrieval module: in response to at least one image to be retrieved uploaded by a user, obtains similar images from the stored detection images based on the image to be retrieved, wherein the similar images are images similar to the image to be retrieved; Defect recognition module: Use the trained defect recognition neural network to perform defect recognition on the inspection image and obtain the defect recognition result.

10. The system according to claim 9, characterized in that The defect recognition result obtained by using the trained defect recognition neural network to perform defect recognition on the detection image includes: Use the trained defect recognition neural network to perform defect recognition on the inspection image to obtain a defect recognition frame list, wherein the defect recognition frame list includes M defect recognition frames arranged in descending order of confidence, and the defect recognition frames in the defect recognition frame list all correspond to the same defect category; Select the defect recognition frame with the highest confidence from the defect recognition frame list for marking, and calculate the intersection-and-union ratio of the defect recognition frame with the highest confidence and the remaining defect recognition frames in the defect recognition frame list, and remove the defect recognition frames whose intersection-and-union ratio is greater than a preset threshold and the defect recognition frame with the highest confidence from the defect recognition frame list; Select the defect recognition frame with the highest confidence from the list of defect recognition frames after elimination for marking, and calculate the intersection and union ratio of the defect recognition frame with the highest confidence and the remaining defect recognition frames in the defect recognition frame list, and eliminate the defect recognition frames whose intersection and union ratio is greater than a preset threshold and the defect recognition frame with the highest confidence from the defect recognition frame list; repeat this step until there is no defect recognition frame in the defect recognition frame list; The marked defect recognition boxes are summarized and output together with the defect categories and confidence levels corresponding to the defect recognition boxes as defect recognition results.

Images