Semi-supervised learning-based welding spot pseudo soldering detection method and related equipment

Abstract

The invention provides a semi-supervised learning-based cold solder joint detection method and related equipment. The method comprises the following steps: constructing a marked welding spot sample data set which is subjected to pseudo soldering defect marking and an unmarked welding spot sample data set which is not subjected to pseudo soldering defect marking; performing feature extraction on each welding spot picture sample of the two data sets; based on the features of the marked welding spot sample data set, training a preset network by using a supervised learning algorithm to obtain an initial welding spot pseudo soldering detection model; detecting the unmarked welding spot sample data set by using the initial welding spot pseudo soldering detection model to obtain a label of each welding spot picture sample in the data set, and forming a new marked welding spot sample data set; using the new marked welding spot sample data set to train the initial welding spot pseudo soldering detection model again to obtain a final welding spot pseudo soldering detection model; and inputting the features corresponding to the to-be-detected welding spot picture into the final welding spot pseudo soldering detection model to obtain a detection result.

Description

Technical Field

[0001] The present invention relates to the technical field of solder joint cold soldering detection, and in particular to a solder joint cold soldering detection method based on semi-supervised learning and related equipment. Background Art

[0002] Solder joint defect is a common circuit fault. The main reasons for the formation of solder joint defect include: (1) Insufficient welding temperature: If the welding temperature is not enough, the solder may not be completely melted, resulting in insufficient melting of the solder joint and a cold solder state. (2) Welding time is too short: If the welding time is too short, the solder may not be fully melted and cannot be fully combined with the base material, resulting in a cold solder defect. (3) Insufficient gas protection: If the gas protection is insufficient, the gas flow at the joint of the wire or the welding machine gun is not smooth, resulting in oxidation or other contamination of the welding interface by the oxygen in the gas, and bubbles may be generated during the entire welding process, resulting in a solder joint defect. (4) Improper adjustment of welding equipment: If the welding equipment is improperly adjusted, such as the current is too small or the voltage is too low, it is difficult for the solder to be completely melted, which eventually leads to a cold solder defect. (5) Improper process operation: Improper welding process operation may also lead to the occurrence of solder joint defect. For example, using too little flux, welding too fast, and not fully cleaning the welding surface may all lead to solder joint defect.

[0003] The main manifestations of cold solder joints are: there will be a circle of obvious collapse around the solder joint, and the solder joint is not smooth, the solder joint color is dark gray, and the solder joint has honeycomb characteristics. Based on this manifestation, the existing methods for detecting cold solder joint defects mainly include: (1) Visual inspection method: Generally, look for components that generate heat first, such as power tubes, high-current diodes, high-power resistors, integrated circuit chips, etc. These components are prone to cold solder joints due to heat. Severe ones can be seen directly, and minor ones can be viewed with a magnifying glass. (2) Current detection method: Check whether the current setting meets the process requirements, and whether the current setting does not increase accordingly when the product load changes, resulting in insufficient current during welding and poor welding. (3) Shaking method: Use your hands or a quiver to shake the low-voltage components one by one to feel whether the components are loose. This is mainly for shaking larger components. (4) Vibration method: When encountering cold solder joints, you can use the knocking method to confirm. Use the handle of a screwdriver to gently knock on the circuit board to determine the location of the cold solder joint. (5) Repair soldering: Repair soldering is a maintenance method used when the fault cannot be found after careful inspection. It is to solder the components within the fault range one by one. In this way, although the real fault point is not found, the maintenance purpose can be achieved. In summary, the current PCB board solder joint cold soldering detection is often carried out through traditional manual detection, which is inefficient. Summary of the invention

[0004] In order to improve the detection efficiency of solder joint cold soldering defects, the present invention provides a solder joint cold soldering detection method and related equipment based on semi-supervised learning, which not only improves the detection efficiency, but also can reduce the detection error rate caused by human factors, and has higher detection accuracy.

[0005] In a first aspect, the present invention provides a method for detecting solder joint defects based on semi-supervised learning, comprising:

[0006] Step 1: Construct a dataset of labeled solder joint samples with marked cold solder joint defects and a dataset of unlabeled solder joint samples without marked cold solder joint defects;

[0007] Step 2: Extract features from each solder joint image sample in the labeled solder joint sample dataset and the unlabeled solder joint sample dataset; wherein the extracted features include the height, volume, and surface quality score of the solder joint;

[0008] Step 3: Based on the features of the labeled solder joint sample data set, the preset network is trained using a supervised learning algorithm to obtain an initial solder joint false solder joint detection model;

[0009] Step 4: Based on the features of the unlabeled solder joint sample data set, the initial solder joint cold joint detection model is used to perform detection, obtain the label of each solder joint image sample in the unlabeled solder joint sample data set, and form a new labeled solder joint sample data set;

[0010] Step 5: retraining the initial solder joint cold solder joint detection model using a new labeled solder joint sample data set to obtain a final solder joint cold solder joint detection model;

[0011] Step 6: Input the features corresponding to the solder joint image to be inspected into the final solder joint cold solder joint detection model to obtain the detection results.

[0012] Furthermore, in step 1, each solder joint image sample in the marked solder joint sample data set or the unmarked solder joint sample data set includes at least three sub-sample images; the at least three sub-sample images are acquired by acquiring images using at least three image acquisition devices evenly distributed around the PCB sample.

[0013] Furthermore, in step 2, feature extraction is performed on each solder joint image sample, specifically including:

[0014] Performing three-dimensional reconstruction using at least three sub-sample images included in the solder joint image sample to obtain a three-dimensional model of the PCB sample, and extracting the height and volume of the solder joint based on the three-dimensional model of the PCB and calibration parameters of the image acquisition device;

[0015] The surface quality score corresponding to the solder joint image sample is extracted by utilizing the preset relationship between the height and volume of the solder joint and the surface quality score.

[0016] Furthermore, it also includes: using a machine learning algorithm to learn a preset relationship between the height and volume of the solder joint and the surface quality score.

[0017] Furthermore, step 3 also includes: normalizing the features of the marked solder joint sample data set; correspondingly, based on the normalized features of the marked solder joint sample data set, using a supervised learning algorithm to train a preset network to obtain an initial solder joint cold solder joint detection model;

[0018] Furthermore, the normalization process includes:

[0019] Select the ideal value of height H i and the ideal value of volume V i According to the selected ideal value, the extracted original height H and original volume V are normalized according to the following formula to obtain the normalized height △H norm and volume △V norm :

[0020]

[0021] And, the extracted original surface quality score S is normalized according to the following formula to obtain the normalized surface quality score △S norm :

[0022] △S norm =(1-S).

[0023] Furthermore, before step 2, it also includes: preprocessing the marked solder joint sample data set and the unmarked solder joint sample data set; the preprocessing includes: one or more operations of graying, equalization, standardization, normalization, image denoising, image enhancement, binarization and morphological processing.

[0024] In a second aspect, the present invention provides a solder joint cold solder joint detection system based on semi-supervised learning, comprising:

[0025] A data set construction module is used to construct a marked solder joint sample data set that has been marked with a cold solder joint defect and an unmarked solder joint sample data set that has not been marked with a cold solder joint defect;

[0026] A feature extraction module is used to extract features from each solder joint image sample in the labeled solder joint sample data set and the unlabeled solder joint sample data set; wherein the extracted features include the height, volume and surface quality score of the solder joint;

[0027] A supervised learning module is used to train a preset network using a supervised learning algorithm based on the features of a labeled solder joint sample data set to obtain an initial solder joint false solder joint detection model;

[0028] A semi-supervised learning module is used to perform detection using the initial solder joint cold solder joint detection model based on the features of the unlabeled solder joint sample data set, obtain the label of each solder joint image sample in the unlabeled solder joint sample data set, and form a new labeled solder joint sample data set; and to use the new labeled solder joint sample data set to train the initial solder joint cold solder joint detection model again to obtain a final solder joint cold solder joint detection model;

[0029] The detection module is used to input the features corresponding to the solder joint image to be detected into the final solder joint cold solder joint detection model to obtain the detection results.

[0030] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in the first aspect when executing the program.

[0031] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0032] Beneficial effects of the present invention:

[0033] The present invention combines supervised learning with semi-supervised learning, and improves the efficiency and accuracy of PCB board solder joint cold soldering detection without the need for a large number of manual annotations of solder joint image samples, greatly reduces the risk of manual detection errors, and ensures the consistency of product quality. In addition, the relevant equipment provided by the present invention has the characteristics of intelligence and automation, and can be linked with other equipment on the production line to improve the production efficiency and product quality of PCB products. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of a flow chart of a solder joint cold solder joint detection method based on semi-supervised learning provided by an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of the collection process of solder joint image samples provided in an embodiment of the present invention;

[0036] Figure 3 A schematic diagram of a process flow of a solder joint cold solder joint detection system based on semi-supervised learning provided by an embodiment of the present invention;

[0037] Figure 4 A structural block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] like Figure 1 As shown, an embodiment of the present invention provides a method for detecting solder joint defects based on semi-supervised learning, comprising:

[0040] S101: constructing a sample data set of marked solder joints that have been marked with cold solder joint defects and a sample data set of unmarked solder joints that have not been marked with cold solder joint defects;

[0041] S102: extracting features from each solder joint image sample in the labeled solder joint sample data set and the unlabeled solder joint sample data set; wherein the extracted features include the height, volume, and surface quality score of the solder joint;

[0042] In order to facilitate feature extraction, in practical applications, it is generally necessary to preprocess the directly acquired original solder joint sample data set; the preprocessing includes: grayscale, equalization, standardization, normalization, image denoising, image enhancement, binarization and morphological processing. Among them, image denoising can maximize the reflection of effective image information; image enhancement is performed through histogram equalization, contrast stretching, etc.; the key threshold of image binarization is determined, and the image background and object are segmented through binarization; when processing binary images, morphological operations (such as corrosion, expansion, opening operation, closing operation) can be applied to improve the shape and structure of the image.

[0043] S103: Based on the features of the labeled solder joint sample data set, a preset network is trained using a supervised learning algorithm to obtain an initial solder joint false solder joint detection model;

[0044] S104: Based on the features of the unlabeled solder joint sample data set, the initial solder joint cold joint detection model is used to perform detection, obtain a label of each solder joint image sample in the unlabeled solder joint sample data set, and form a new labeled solder joint sample data set;

[0045] S105: retraining the initial solder joint cold solder joint detection model using a new labeled solder joint sample data set to obtain a final solder joint cold solder joint detection model;

[0046] S106: Input the features corresponding to the image of the solder joint to be inspected into the final solder joint cold solder joint inspection model to obtain the inspection result.

[0047] Due to the high cost of data annotation, it is generally impossible to obtain complete labels for all solder joint image sample points. Therefore, the embodiment of the present invention makes full use of limited labeled data to train an initial solder joint cold welding detection model (hereinafter referred to as the initial model), and then based on the principle of semi-supervised learning, uses the initial model to detect the unlabeled solder joint sample data set to obtain a new labeled solder joint sample data set, thereby training the initial model to obtain the final solder joint cold welding detection model. Compared with the existing solder joint cold welding defect detection method, the present invention can improve the detection efficiency and detection accuracy.

[0048] In one embodiment, in order to improve the quality of the collected images and facilitate the subsequent feature extraction process, such as Figure 2 As shown, in the process of constructing the data set, three image acquisition devices are used to simultaneously take pictures of the PCB sample from positions 120 degrees apart. In addition, in order to minimize the interference of ambient light during the shooting process, in practical applications, the PCB sample can generally be placed under a parallel linear array light source for collection. In this way, each solder joint image sample in the marked solder joint sample data set or the unmarked solder joint sample data set in this embodiment includes at least three sub-sample images.

[0049] It can be understood that this image acquisition method can provide three-dimensional information of the PCB sample. By taking pictures from three different angles and using image processing and computer vision technology, a three-dimensional model of the PCB can be reconstructed. This model can provide information about the size, shape and surface features of the PCB.

[0050] In one embodiment, the height, volume and surface quality score of the solder joint in each solder joint image sample are extracted in the following manners.

[0051] S201: Performing three-dimensional reconstruction using at least three sub-sample images included in the solder joint image sample to obtain a three-dimensional model of the PCB sample, and extracting the height and volume of the solder joint based on the three-dimensional model of the PCB and calibration parameters of an image acquisition device;

[0052] Specifically, since the three sub-sample images are view images of the PCB sample collected from different perspectives, the three-dimensional model of the PCB can be restored from these different view images using image matching and three-dimensional reconstruction algorithms. Based on the three-dimensional model of the PCB, the volume of the PCB can be calculated using voxelization or integration methods, and the volume of the solder joint can naturally be obtained.

[0053] In this embodiment, the height measurement is performed using stereo vision technology. The key feature points or areas for measuring height are identified from the customized main view through image processing algorithms such as edge detection and threshold segmentation, and then the corresponding matching points of these feature points in the left and right views are found from the other two views. By calculating the parallax between these matching points and combining the calibration parameters of the camera (such as focal length, camera internal parameters, etc.), the coordinates of the key feature points in three-dimensional space can be accurately calculated, where the Z-axis (i.e., vertical direction) coordinate represents the height of the key feature point, that is, the height of the detected solder joint.

[0054] S202: Utilizing a preset relationship between the height and volume of a solder joint and the surface quality score, extracting a surface quality score corresponding to the solder joint image sample.

[0055] In this embodiment, PCB images with marked surface quality scores are prepared in advance, and based on these PCB images, a machine learning algorithm is used to learn the preset relationship between the height and volume of the solder joints in the image and the surface quality score. Finally, when the model training is completed, a new PCB image is input into the model, and the model will output a surface quality score between 0 and 1.

[0056] Since there are inconsistent dimensions between the extracted solder joint features (for example, the unit of solder joint height H may be millimeters, the unit of solder joint volume V may be cubic millimeters, and the surface quality score S is a dimensionless value, which may range from 0 to 1, where 1 represents perfect surface quality), it is not conducive to model training. Therefore, in one embodiment, in the above-mentioned step S103, it also includes: normalizing the features of the labeled solder joint sample data set; correspondingly, based on the normalized features of the labeled solder joint sample data set, using a supervised learning algorithm to train the preset network to obtain an initial solder joint cold solder joint detection model.

[0057] It is understandable that the preset network may select an existing network architecture, or may customize the network structure as needed, which is not limited in this embodiment.

[0058] As an implementable method, the extracted original height, original volume and original surface quality score of the solder joint are normalized in the following manners.

[0059] Select the ideal value of height H i and the ideal value of volume V i According to the selected ideal value, the extracted original height H and original volume V are normalized according to the following formula to obtain the normalized height △H norm and volume △V norm :

[0060]

[0061] And, the extracted original surface quality score S is normalized according to the following formula to obtain the normalized surface quality score △S norm :

[0062] △S norm =(1-S).

[0063] like Figure 3 As shown, an embodiment of the present invention provides a solder joint cold solder joint detection system based on semi-supervised learning, including: a data set construction module, a feature extraction module, a supervised learning module, a semi-supervised learning module and a detection module.

[0064] Among them, the data set construction module is used to construct a labeled solder joint sample data set that has been marked with cold solder joint defects and an unlabeled solder joint sample data set that has not been marked with cold solder joint defects; the feature extraction module is used to extract features from each solder joint image sample in the labeled solder joint sample data set and the unlabeled solder joint sample data set; wherein the extracted features include the height, volume and surface quality score of the solder joint; the supervised learning module is used to train the preset network using a supervised learning algorithm based on the features of the labeled solder joint sample data set to obtain an initial solder joint cold solder joint detection model; the semi-supervised learning module is used to perform detection based on the features of the unlabeled solder joint sample data set using the initial solder joint cold solder joint detection model to obtain the label of each solder joint image sample in the unlabeled solder joint sample data set and form a new labeled solder joint sample data set; and it is used to train the initial solder joint cold solder joint detection model again using the new labeled solder joint sample data set to obtain a final solder joint cold solder joint detection model; the detection module is used to input the features corresponding to the solder joint image to be detected into the final solder joint cold solder joint detection model to obtain the detection results.

[0065] It should be noted that the system provided in the embodiment of the present invention is for implementing the above method, and its specific functions can be referred to the above method embodiments, which will not be described again here.

[0066] The solder joint cold solder joint detection system based on semi-supervised learning provided by the embodiment of the present invention has higher detection accuracy and efficiency through the mutual cooperation of semi-supervised and supervised learning, and can also reduce the influence of human factors on the detection results.

[0067] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4As shown, the electronic device may include: a processor (processor) 401, a communication interface (Communications Interface) 402, a memory (memory) 403 and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404. The processor 401 can call the logic instructions in the memory 403 to execute the solder joint cold welding detection method, which includes: step 1: constructing a marked solder joint sample data set that has been marked with cold welding defects and an unmarked solder joint sample data set that has not been marked with cold welding defects; step 2: extracting features from each solder joint image sample in the marked solder joint sample data set and the unmarked solder joint sample data set; wherein the extracted features include the height, volume and surface quality score of the solder joint; step 3: based on the features of the marked solder joint sample data set, using a supervised learning algorithm to train a preset network to obtain an initial solder joint cold welding detection model; step 4: based on the features of the unmarked solder joint sample data set, using the initial solder joint cold welding detection model to perform detection, obtain the label of each solder joint image sample in the unmarked solder joint sample data set, and form a new marked solder joint sample data set; step 5: using the new marked solder joint sample data set to train the initial solder joint cold welding detection model again to obtain a final solder joint cold welding detection model; step 6: inputting the features corresponding to the solder joint image to be detected into the final solder joint cold welding detection model to obtain the detection result.

[0068] In addition, when the logic instructions in the above-mentioned memory 403 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0069] The embodiment of the present invention further provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions are executed by a computer, the computer can execute the methods provided by the above-mentioned method embodiments, for example, including: step 1: constructing a marked solder joint sample data set that has been marked with a cold solder joint defect and an unmarked solder joint sample data set that has not been marked with a cold solder joint defect; step 2: extracting features from each solder joint image sample in the marked solder joint sample data set and the unmarked solder joint sample data set; wherein the extracted features include the height, volume and surface quality evaluation of the solder joint Step 3: Based on the characteristics of the labeled solder joint sample data set, a supervised learning algorithm is used to train the preset network to obtain an initial solder joint cold solder joint detection model; Step 4: Based on the characteristics of the unlabeled solder joint sample data set, the initial solder joint cold solder joint detection model is used for detection to obtain the label of each solder joint image sample in the unlabeled solder joint sample data set, and a new labeled solder joint sample data set is formed; Step 5: The initial solder joint cold solder joint detection model is trained again using the new labeled solder joint sample data set to obtain a final solder joint cold solder joint detection model; Step 6: The features corresponding to the solder joint image to be detected are input into the final solder joint cold solder joint detection model to obtain the detection results.

[0070] The embodiment of the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method provided by the above-mentioned method embodiments is implemented, for example, including: step 1: constructing a marked solder joint sample data set that has been marked with a cold solder joint defect and an unmarked solder joint sample data set that has not been marked with a cold solder joint defect; step 2: extracting features from each solder joint image sample in the marked solder joint sample data set and the unmarked solder joint sample data set; wherein the extracted features include the height, volume and surface quality score of the solder joint; step 3: based on the marked solder joint sample data set Features, use a supervised learning algorithm to train the preset network to obtain an initial solder joint cold solder joint detection model; Step 4: Based on the features of the unlabeled solder joint sample data set, use the initial solder joint cold solder joint detection model to perform detection, obtain the label of each solder joint image sample in the unlabeled solder joint sample data set, and form a new labeled solder joint sample data set; Step 5: Use the new labeled solder joint sample data set to train the initial solder joint cold solder joint detection model again to obtain the final solder joint cold solder joint detection model; Step 6: Input the features corresponding to the solder joint image to be detected into the final solder joint cold solder joint detection model to obtain the detection results.

[0071] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting solder joint defects based on semi-supervised learning, characterized in that: include: Step 1: Construct a dataset of labeled solder joint samples with marked cold solder joint defects and a dataset of unlabeled solder joint samples without marked cold solder joint defects; Step 2: Extract features from each solder joint image sample in the labeled solder joint sample dataset and the unlabeled solder joint sample dataset; wherein the extracted features include the height, volume, and surface quality score of the solder joint; Step 3: Based on the features of the labeled solder joint sample data set, the preset network is trained using a supervised learning algorithm to obtain an initial solder joint false solder joint detection model; Step 4: Based on the features of the unlabeled solder joint sample data set, the initial solder joint cold joint detection model is used to perform detection, obtain the label of each solder joint image sample in the unlabeled solder joint sample data set, and form a new labeled solder joint sample data set; Step 5: retraining the initial solder joint cold solder joint detection model using a new labeled solder joint sample data set to obtain a final solder joint cold solder joint detection model; Step 6: Input the features corresponding to the solder joint image to be inspected into the final solder joint cold solder joint detection model to obtain the detection results.

2. The method for detecting solder joint defects based on semi-supervised learning according to claim 1, characterized in that: In step 1, each solder joint image sample in the marked solder joint sample data set or the unmarked solder joint sample data set includes at least three sub-sample images; the at least three sub-sample images are acquired by using at least three image acquisition devices evenly distributed around the PCB sample.

3. The method for detecting solder joint defects based on semi-supervised learning according to claim 2, characterized in that: In step 2, feature extraction is performed on each solder joint image sample, specifically including: Performing three-dimensional reconstruction using at least three sub-sample images included in the solder joint image sample to obtain a three-dimensional model of the PCB sample, and extracting the height and volume of the solder joint based on the three-dimensional model of the PCB and calibration parameters of the image acquisition device; The surface quality score corresponding to the solder joint image sample is extracted by utilizing the preset relationship between the height and volume of the solder joint and the surface quality score.

4. The method for detecting solder joint defects based on semi-supervised learning according to claim 3, characterized in that: It also includes: using a machine learning algorithm to learn a preset relationship between the height and volume of the solder joint and the surface quality score.

5. The method for detecting solder joint defects based on semi-supervised learning according to claim 3, characterized in that: Step 3 also includes: normalizing the features of the marked solder joint sample data set; correspondingly, based on the normalized features of the marked solder joint sample data set, a supervised learning algorithm is used to train the preset network to obtain an initial solder joint cold solder joint detection model.

6. The method for detecting solder joint defects based on semi-supervised learning according to claim 5, characterized in that: The normalization process includes: Select the ideal value of height H i and the ideal value of volume V i According to the selected ideal value, the extracted original height H and original volume V are normalized according to the following formula to obtain the normalized height △H norm and volume △V norm : And, the extracted original surface quality score S is normalized according to the following formula to obtain the normalized surface quality score △S norm : △S norm =(1-S)。 7. The method for detecting solder joint defects based on semi-supervised learning according to claim 1, characterized in that: Before step 2, the method further includes: preprocessing the marked solder joint sample data set and the unmarked solder joint sample data set; the preprocessing includes: one or more operations of graying, equalization, standardization, normalization, image denoising, image enhancement, binarization and morphological processing.

8. A solder joint cold solder joint detection system based on semi-supervised learning, characterized in that: include: A data set construction module is used to construct a marked solder joint sample data set that has been marked with a cold solder joint defect and an unmarked solder joint sample data set that has not been marked with a cold solder joint defect; A feature extraction module is used to extract features from each solder joint image sample in the labeled solder joint sample data set and the unlabeled solder joint sample data set; wherein the extracted features include the height, volume and surface quality score of the solder joint; A supervised learning module is used to train a preset network using a supervised learning algorithm based on the features of a labeled solder joint sample data set to obtain an initial solder joint false solder joint detection model; A semi-supervised learning module is used to perform detection using the initial solder joint cold solder joint detection model based on the features of the unlabeled solder joint sample data set, obtain the label of each solder joint image sample in the unlabeled solder joint sample data set, and form a new labeled solder joint sample data set; and to use the new labeled solder joint sample data set to train the initial solder joint cold solder joint detection model again to obtain a final solder joint cold solder joint detection model; The detection module is used to input the features corresponding to the solder joint image to be detected into the final solder joint cold solder joint detection model to obtain the detection results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images