An industrial defect detection optimization method, system, device, equipment and storage medium

Abstract

The application discloses an industrial defect detection optimization method, system, device, equipment and storage medium, the method comprises the following steps: acquiring an original image, labeling defect sample data in the original image; enhancing the defect sample data to obtain an original data set; cloud online learning is carried out according to the original data set, and a defect detection model is trained; the defect detection model is applied to detect industrial product samples, and detection results are obtained; a plurality of detection results are randomly selected, whether the detection of the defect detection model on the industrial product samples is correct is judged, when the detection result is correct, the defect detection model is continuously used to detect the industrial product samples, otherwise, the defect sample data in the original image is labeled again. When the model detection result deviates, improvement can be made, the newly appeared effective sample information in the detection process is fully utilized, the performance of the model does not continuously decrease in the application process, and the detection performance is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of industrial defect detection, and relates to an industrial defect detection optimization method, system, device, equipment and storage medium. BACKGROUND

[0002] Defect detection is an important link in the industrial production process. The traditional defect detection scheme based on manual detection has low detection efficiency and has the risk of missing detection. In recent years, target detection models based on deep learning have been gradually applied to industrial defect detection, realizing semi-automation of industrial defect detection and improving detection efficiency to a certain extent. Defect detection systems based on deep learning have gradually attracted the attention of relevant practitioners, and different forms of defect detection systems have emerged. Since the existing target detection models based on deep learning require a large amount of training data, in actual application, a large amount of sample data often needs to be collected first, the model needs to be trained, and the model meeting the requirements needs to be manually screened out, so that the model can be deployed in the actual industrial defect detection system, and the full-process automation operation cannot be realized. The existing industrial defect detection system often defaults that the detection accuracy of the deployed detection model always meets the requirements. However, due to various uncertain factors, the detection accuracy of the model often decreases to an unacceptable level during the actual operation of the system, and the performance evaluation during the operation cannot be realized. The existing defect detection system based on deep learning often adopts the mode of "one-time deployment and lifelong use", and does not use the effective samples newly appeared in the running process of the system. SUMMARY

[0003] The purpose of the present application is to solve the problem that the defect detection system in the prior art cannot timely find the decrease of the detection accuracy of the model in the actual operation, and to provide an industrial defect detection optimization method, system, device, equipment and storage medium.

[0004] To achieve the above purpose, the following technical solutions are adopted:

[0005] An industrial defect detection optimization method comprises the following steps:

[0006] An original image is acquired, and defect sample data in the original image is labeled;

[0007] The defect sample data is enhanced to obtain an original data set;

[0008] Cloud online learning is performed according to the original data set, and a defect detection model is trained;

[0009] The defect detection model is applied to detect industrial product samples to obtain a detection result;

[0010] Randomly select several test results, judge whether the defect detection model is correct for the detection of the industrial product sample, if the test result is correct, continue to use the defect detection model to detect the industrial product sample, otherwise, re-label the defect sample data in the original image.

[0011] Further improvement of the application is:

[0012] The defect sample data enhancement includes the following steps:

[0013] A random number in [0, 1] is obtained as a gamma transformation parameter, and the original image is subjected to gamma transformation;

[0014] The original image is rotated at a random angle around the center point of the original image;

[0015] A random number r in [0.7, 1] is obtained as a scale parameter, and two independent random numbers a and b in [0, 1] are obtained as position parameters;

[0016] In the rectangular region with (a, b) and (min(a+r, 1), min(b+r, 1)) as opposite angles in the original image, random cropping is performed.

[0017] After the cloud online learning according to the original data set is trained to obtain the defect detection model, the detection accuracy of the defect detection model is tested, and after the detection accuracy is qualified, the defect detection model is saved; if the detection accuracy is unqualified, the cloud online learning is continued.

[0018] An industrial defect detection optimization system comprises:

[0019] The data acquisition module is used to acquire an original image and label defect sample data in the original image.

[0020] The data enhancement module is used to enhance the defect sample data to obtain an original data set.

[0021] The cloud server is used to perform cloud online learning according to the original data set to train a defect detection model.

[0022] The detection module is used to detect an industrial product sample according to the defect detection model to obtain a test result.

[0023] The result verification module is used to randomly select several test results, judge whether the defect detection model is correct for the detection of the industrial product sample, if the test result is correct, continue to use the defect detection model to detect the industrial product sample, otherwise, re-label the defect sample data in the original image.

[0024] An industrial defect detection optimization device, comprising:

[0025] A client is configured to collect original images, pre-process the original images to obtain an original data set, transmit the original images to an inference server, transmit the original data set to an online learning end, and verify the detection result of the inference server;

[0026] An online learning end is configured to perform cloud online learning according to the original data set, train a defect detection model, and deliver the defect detection model to the inference server;

[0027] An inference server is configured to receive the original images, call the defect detection model, and send the detection result to the client.

[0028] The client comprises an image acquisition unit, an image pre-processing unit, a result visualization unit, and an artificial re-detection unit;

[0029] The image acquisition unit acquires original images using an industrial camera, transmits the original images to the image pre-processing unit for enhancement to obtain an original data set, and the result visualization unit is configured to receive the detection result of the inference server and verify the detection result by the artificial re-detection unit.

[0030] The online learning end comprises a first request management unit, a dynamic database unit, a model training unit, and a model testing unit;

[0031] The first request management unit receives the original data set transmitted by the client, stores the original data set in the dynamic database unit, trains the data in the dynamic database unit by the model training unit to obtain a defect detection model, and transmits the defect detection model to the inference server after testing the detection accuracy by the model testing unit.

[0032] The inference server comprises a second request management unit, a version management unit, and a model loading unit;

[0033] The second request management unit receives the original images transmitted by the client, calls the defect detection model from the version management unit, loads the called defect detection model by the model loading unit, detects the original images, and sends the detection result to the client.

[0034] A device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of the preceding items when executing the computer program.

[0035] A computer readable storage medium stores a computer program, the computer program is executed by a processor to implement the steps of the method of any one of the preceding.

[0036] Compared with the prior art, the present application has the following beneficial effects:

[0037] In the application process of the defect detection model, the detection results are sampled, when the detection results are correct, the model is continuously used for detection, when the detection results are incorrect, the defect sample data is re-labeled and online learning in the cloud is performed, when the model detection result deviates, improvement can be made, the effective sample information newly appeared in the detection process is fully utilized, and it is ensured that the model will not continuously decrease in performance in the application process, and the detection performance is improved. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments, and it should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation to the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0039] Figure 1 The present application is an industrial defect detection optimization method flowchart;

[0040] Figure 2 The present application is an industrial defect detection optimization method flowchart;

[0041] Figure 3 The present application is an industrial defect detection optimization system flowchart;

[0042] Figure 4 The present application is an industrial defect detection optimization device structure schematic diagram;

[0043] Figure 5 The present application is a client specific structure schematic diagram;

[0044] Figure 6 The present application is an online learning end specific structure schematic diagram;

[0045] Figure 7 The present application is a reasoning server specific structure schematic diagram. DETAILED DESCRIPTION

[0046] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0047] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0048] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0049] In the description of the embodiments of the present application, it should be noted that if the terms "upper", "lower", "horizontal", "inner" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship when the product of the present application is used, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second" and the like are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0050] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly inclined. For example, "horizontal" only means that its direction is relatively more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly inclined.

[0051] In the description of the embodiments of the present application, it should also be noted that unless otherwise explicitly specified and limited, if the terms "arrangement", "installation", "connection", "connection" appear, they should be understood in a broad sense, for example, they can be fixedly connected, or detachably connected, or integrally connected; can be mechanically connected, or electrically connected; can be directly connected, or indirectly connected through an intermediate medium, or the communication inside two elements. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0052] The present application will be described in further detail below with reference to the drawings:

[0053] Referring toFigure 1 , as an industrial defect detection optimization method flow chart of the application, comprising the following steps:

[0054] S1, obtaining the original image, labeling the defect sample data in the original image;

[0055] S2, the defect sample data is enhanced, and the original data set is obtained;

[0056] S2.1, a random number in [0, 1] is obtained as a gamma transform parameter, and the original image is subjected to gamma transform;

[0057] S2.2, around the center point of the original image, the original image is rotated at a random angle;

[0058] S2.3, a random number r in [0.7, 1] is obtained as a scale parameter, and two independent random numbers a and b in [0, 1] are obtained as position parameters;

[0059] S2.4, in the original image, in the rectangular region with (a, b) point and (min(a+r, 1), min(b+r, 1)) point as opposite angles, random cutting is carried out.

[0060] S3, according to the original data set, cloud online learning is carried out, and the defect detection model is trained;

[0061] S4, according to the defect detection model, the industrial product sample is detected, and the detection result is obtained;

[0062] S5, randomly select several detection results, judge whether the detection result of the defect detection model to the industrial product sample is correct, if the detection result is correct, continue to use the defect detection model to detect the industrial product sample, otherwise, re-label the defect sample data in the original image.

[0063] Reference Figure 2 , as an industrial defect detection optimization method flow chart of the application, comprising the following steps:

[0064] S1, system initialization, enter "manual detection" mode, label the defect sample data appeared;

[0065] S2, when the number of labeled defect sample data reaches the set size, the defect sample data is enhanced, and the original data set is obtained;

[0066] S3, based on the original data set, cloud online learning is carried out, and the defect detection model is trained;

[0067] S4, test the defect detection model, when the test result is higher than the set standard, enter the subsequent step, apply and deploy the defect detection model, and detect the industrial product sample; otherwise, return to S1, and continue to collect effective defect sample information;

[0068] S5, when detecting the industrial product sample, save the detected sample information, and when the detected sample information reaches a certain scale, randomly select a plurality of pictures therefrom, and artificially judge whether the model detection result is correct;

[0069] S6, when the model detection result is correct, continue to apply the defect detection model to detect the industrial product sample; otherwise, the system issues an alarm signal, and returns to S1, and continues to collect effective defect sample information.

[0070] When the defect sample data is enhanced in S2, the following two data enhancement strategies are adopted. One kind of data enhancement strategy is similar to the traditional data enhancement strategy: a random number is selected in the interval [0, 1] as the parameter in the image gamma transformation, and the original image is subjected to gamma transformation; then, the original image is rotated at a random angle around the center point of the original image; then, a random number r is selected from [0.7, 1] as a scale parameter, and two independent random numbers a and b are selected from [0, 1] as position parameters, and the original image is randomly cropped, and the region is a rectangle region with opposite corners at (a, b) and (min(a+r, 1), min(b+r, 1)) of the original image. Another kind of data enhancement strategy is to randomly copy the image of the defect region in the existing defect sample to the sample image without defects.

[0071] In S5, the detected sample information includes two types: one is the defect-free sample detected by the model, and the other is the defective sample detected by the model. The proportion between the two is kept unchanged, and a plurality of pictures are randomly selected from the detected samples to artificially determine whether the model detection result is correct.

[0072] Referring to Figure 3 , it is an industrial defect detection optimization system schematic diagram, comprising

[0073] The data acquisition module is used for acquiring an original image and labeling defect sample data in the original image;

[0074] The data enhancement module is used for enhancing the defect sample data to obtain an original data set;

[0075] The cloud server is used for cloud online learning according to the original data set, and training a defect detection model;

[0076] The detection module is configured to detect the industrial product sample according to the defect detection model to obtain a detection result.

[0077] The result verification module is configured to randomly select a plurality of detection results, judge whether the detection of the industrial product sample by the defect detection model is correct, continue to detect the industrial product sample by using the defect detection model when the detection result is correct, or re-label the defect sample data in the original image otherwise.

[0078] Referring to Figure 4 An industrial defect detection optimization device structure schematic diagram is provided, which comprises:

[0079] The client is configured to collect an original image, pre-process the original image to obtain an original data set, transmit the original image to an inference server, transmit the original data set to an online learning end, and verify a detection result of the inference server.

[0080] The online learning end is configured to perform cloud online learning according to the original data set, train a defect detection model, and deliver the defect detection model to the inference server.

[0081] The inference server is configured to receive the original image, call the defect detection model, and send a detection result to the client.

[0082] Referring to Figure 5 A client specific structure schematic diagram is provided, which comprises an image collection unit, an image pre-processing unit, a result visualization unit, and a manual re-detection unit. The image collection unit collects an original image by using an industrial camera, transmits the original image to the image pre-processing unit for enhancement to obtain an original data set, the result visualization unit is configured to receive a detection result of the inference server, and the manual re-detection unit is configured to verify the detection result.

[0083] The image acquisition unit comprises a light source, an industrial camera and a supporting device. The industrial camera is placed above the industrial product to be detected by the supporting device to acquire the image of the product on the production line. The light source is used to improve the imaging environment of the industrial camera, so that the image acquired by the industrial camera can clearly present the defects on the surface of the product. The supporting device is used to fix the industrial camera and the light source, so as to ensure that the industrial camera can clearly and completely acquire the image of the product. In the embodiment, the industrial camera adopts a CMOS area array imaging camera, the optical resolution of which is 5120*5120, the frame rate is 15.1 fps, the pixel size is 2.5um*2.5um, the pixel depth is 8bit, and the exposure time is 10us-1s. The light source adopts an area array light source, the size of which is 0.7*0.7m, the power input voltage is AC90-264V, the power input frequency range is 47-63Hz, the output current adjustment range is 0-8A, the brightness adjustment method is manual button adjustment or PC remote adjustment, the brightness adjustment mode is constant current type, and the brightness adjustment level is 256 levels. The image preprocessing unit is a gamma enhancement subprogram, which performs gamma enhancement on the defect sample data, the coefficient of the gamma enhancement is set to 0.7, and the imaging effect is improved. The result visualization unit comprises a subprogram for visualizing the detection result output by the inference service end, marking a box at the corresponding position of the original image, and distinguishing different types of defects by different colors. The defect types include three types: scratch (scratch), spot (groove) and rust (rust). The artificial re-detection unit comprises a subprogram for artificially re-labeling the detection result. Based on the output of the result visualization unit, whether the defect detection model has false detection / miss detection is judged by the artificial. If there is no false detection / miss detection, a submit button is directly clicked to save the final detection result. If there is false detection / miss detection, the false detection is corrected and the missed detection is re-labeled, and finally the corrected defect information is saved. The final defect information is saved to a text document. An independent text document contains all defect information of an industrial product image. Each line of the text document corresponds to a defect, and specifically contains the type of the defect and the position of the defect. The naming of the file document is consistent with the naming of the corresponding industrial product image file collected.

[0084] The specific working steps of the client are:

[0085] S1, a subprogram for controlling the camera to shoot the image in the Daheng image SDK is called to acquire the original image;

[0086] S2, the original image is transmitted to the cloud server and stored in the original image database, a defect detection model is called to perform defect detection, and the detection result is saved to the model detection result database;

[0087] S3, judge whether the sample quantity detected by the model reaches the set scale, when reaching the set scale, extract a number of samples from the model detection result database, artificially judge whether the detection result of the defect detection model is correct, if correct, continue to detect, otherwise, enter the "manual detection" mode.

[0088] Referring to Figure 6 The specific structure diagram of the online learning end in the application is shown, the online learning end includes a first request management unit, a dynamic database unit, a model training unit and a model testing unit; the first request management unit receives the original data set transmitted by the client, stores the original data set to the dynamic database unit, trains the data in the dynamic database unit by the model training unit to obtain a defect detection model, and transmits the defect detection model to the inference service end after testing the detection precision by the model testing unit.

[0089] The specific working steps of the online learning end are as follows:

[0090] S1, judge whether the data received by the data transmission interface is legal, and exclude data loss caused by network and the like. When the data is legal, add the data to the dynamic database unit, otherwise, discard the data;

[0091] S2, judge whether the data quantity in the dynamic database unit reaches a certain scale, when reaching a certain scale, perform online training of the model, otherwise, continue to receive data from the data transmission interface;

[0092] S3, test the model obtained by training, when the test result reaches the set standard, save the defect detection model and transmit it to the inference service end for application of the model; otherwise, continue to receive data from the data transmission interface.

[0093] Referring to Figure 7 The specific structure diagram of the inference service end in the application is shown, the inference service end includes a second request management unit, a version management unit and a model loading unit; the second request management unit receives the original image transmitted by the client, calls the defect detection model from the version management unit, loads the called defect detection model by the model loading unit, detects the original image, and sends the detection result to the client.

[0094] The specific working steps of the inference service end are as follows:

[0095] S1, the request management unit receives the image data to be processed, analyzes the data to obtain the image to be processed, the called detection model and the priority information and the like;

[0096] S2, the version management unit queries whether the detection model called by the request has been started, and if yes, directly calls the detection model of the version to process; otherwise, searches the model of the version, loads the detection model by the model loading unit, and detects the image to be processed;

[0097] S3, the inference server sends the result to the client by the data transmission interface.

[0098] The online learning end and the inference server are constructed based on a TorchServe framework of Pytorch, and the specific steps are as follows:

[0099] S1, according to the guide provided by TorchServe, the model file, the weight file and the operation handle file are obtained;

[0100] S2, the model file, the weight file and the operation handle file are packaged into a MAR file by using the scaffolding tool provided by TorchServe;

[0101] S3, the obtained MAR file is encapsulated into a service with a RESTful API interface by using the scaffolding tool provided by TorchServe.

[0102] An embodiment of the present application provides a terminal device. The terminal device of the embodiment comprises a processor, a memory and a computer program stored in the memory and executable on the processor. The processor implements the steps in each of the industrial defect detection method embodiments when executing the computer program. Alternatively, the processor implements the functions of each module / unit in each of the device embodiments when executing the computer program.

[0103] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application.

[0104] The industrial defect detection device / terminal device can be a desktop computer, a notebook computer, a palm computer and a cloud server, etc. The industrial defect detection device / terminal device can include, but is not limited to, a processor and a memory.

[0105] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc.

[0106] The memory can be used to store the computer programs and / or modules, and the processor realizes various functions of the industrial defect detection device / terminal equipment by running or executing the computer programs and / or modules stored in the memory, and calling data stored in the memory.

[0107] The modules / units integrated in the industrial defect detection device / terminal equipment, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can realize the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content included in the computer readable 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, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0108] The above only describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An industrial defect detection optimization method, characterized by, The method comprises the following steps: Obtaining an original image, and labeling defect sample data in the original image; Enhancing the defect sample data to obtain an original data set; The defect sample data enhancement comprises the following steps: Obtaining a random number in [0, 1] as a gamma transformation parameter, and performing gamma transformation on the original image; Rotating the original image at a random angle around the center point of the original image; Obtaining a random number r in [0.7, 1] as a scale parameter, and obtaining two independent random numbers a and b in [0, 1] as position parameters; Randomly cropping a rectangular region with opposite corners at (a, b) and (min(a+r, 1), min(b+r, 1)) in the original image; Performing cloud online learning according to the original data set to train a defect detection model; after the cloud online learning according to the original data set to train the defect detection model, the method further comprises testing the detection accuracy of the defect detection model, saving the defect detection model when the detection accuracy is qualified, and continuing the cloud online learning when the detection accuracy is unqualified; Applying the defect detection model to detect industrial product samples to obtain detection results; Randomly selecting a plurality of detection results to determine whether the defect detection model correctly detects the industrial product samples; when the detection results are correct, continuing to detect the industrial product samples by using the defect detection model; otherwise, re-labeling the defect sample data in the original image; The industrial defect detection optimization device comprises: A client, which is configured to collect an original image, pre-process the original image to obtain an original data set, transmit the original image to an inference server, transmit the original data set to an online learning end, and verify a detection result of the inference server; the client comprises an image collection unit, an image pre-processing unit, a result visualization unit, and a manual re-detection unit; the image collection unit collects the original image by using an industrial camera, transmits the original image to the image pre-processing unit for enhancement to obtain the original data set, and the result visualization unit is configured to receive the detection result of the inference server and verify the detection result by the manual re-detection unit; The online learning end performs cloud online learning according to the original data set to train a defect detection model, and transmits the defect detection model to the inference server; the online learning end comprises a first request management unit, a dynamic database unit, a model training unit, and a model testing unit; the first request management unit receives the original data set transmitted by the client, stores the original data set in the dynamic database unit, trains data in the dynamic database unit by the model training unit to obtain the defect detection model, and transmits the defect detection model to the inference server after testing the detection accuracy by the model testing unit; The inference server receives the original image, calls the defect detection model, and sends the detection result to the client. The inference server includes a second request management unit, a version management unit, and a model loading unit. The second request management unit receives the original image transmitted by the client, calls the defect detection model from the version management unit, loads the called defect detection model by the model loading unit, detects the original image, and sends the detection result to the client.

2. An industrial defect detection optimization system using the industrial defect detection optimization method of claim 1, characterized by, include: The data acquisition module is used to acquire the original image and annotate the defect sample data in the original image. A data augmentation module is used to augment defective sample data to obtain the original dataset; A cloud server, which is used to perform online learning based on the original dataset to train a defect detection model; The detection module is used to detect industrial product samples according to a defect detection model and obtain detection results; The result verification module is used to randomly select several detection results to determine whether the defect detection model is correct in detecting industrial product samples. If the detection result is correct, the defect detection model is used to continue to detect industrial product samples; otherwise, the defect sample data in the original image is re-labeled.

3. An apparatus for optimizing industrial defect detection, using the method for optimizing industrial defect detection according to claim 1, characterized in that, include: The client is used to acquire raw images, preprocess them to obtain a raw dataset, transmit the raw images to the inference server, transmit the raw dataset to the online learning terminal, and verify the detection results of the inference server. The client includes an image acquisition unit, an image preprocessing unit, a result visualization unit, and a manual re-detection unit. The image acquisition unit uses an industrial camera to acquire raw images, transmits the raw images to the image preprocessing unit for enhancement, and obtains the raw dataset. The result visualization unit receives the detection results from the inference server, and the manual re-detection unit verifies the detection results. The online learning terminal trains a defect detection model using the original dataset through cloud-based online learning, and then transmits the defect detection model to the inference server. The online learning terminal includes a first request management unit, a dynamic database unit, a model training unit, and a model testing unit. The first request management unit receives the original dataset transmitted by the client, stores the original dataset in the dynamic database unit, trains the model on the data in the dynamic database unit to obtain the defect detection model, and after the model testing unit tests the detection accuracy, the defect detection model is transmitted to the inference server. The inference server receives the original image, calls the defect detection model, and sends the detection result to the client; the inference server comprises a second request management unit, a version management unit and a model loading unit; the second request management unit receives the original image transmitted by the client, calls the defect detection model from the version management unit, loads the called defect detection model by the model loading unit, detects the original image, and sends the detection result to the client.

4. An apparatus comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, The computer program is executed by the processor to implement the steps of the method of claim 1.

5. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 4. The computer program is executed by the processor to implement the steps of the method of claim 1.

Images