Half-hole burr detection method and system for acid etching forming of AIOT module board

Abstract

The invention relates to the field of intelligent detection, and provides a half-hole burr detection method and system for acid etching forming of an AIOT module board, and the method comprises the steps: firstly, carrying out rough fishing slotting at a preset position of the AIOT module board by using a common milling cutter to form a rough fishing slotting region, then milling the rough fishing slotting region along a left compensation path by using a reverse milling cutter, and carrying out rough fishing slotting on the rough fishing slotting region; and milling the rough-scooping slotting area along the right compensation route by using the common milling cutter, then milling the rough-scooping slotting area along the left compensation route by using the reverse milling cutter so as to obtain a sunken half hole at a preset position, and finally carrying out semantic space strengthening analysis based on detection-standard image edge features on the half hole so as to obtain a burr detection result. In this way, the problem of burrs generated during acid etching forming of the AIoT module board can be effectively controlled, and then improvement of the process reliability is facilitated.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to a method and system for detecting burrs in half-holes for acid etching forming of AIoT module boards. Background Art

[0002] At present, with the rapid development of technology, AIoT module products, by integrating the Internet of Things and artificial intelligence, have brought profound changes to various industries. Their applications are extremely extensive. In the industrial field, they help traditional manufacturing industries transform into intelligent manufacturing. By real-time monitoring equipment status and optimizing production processes, they effectively improve production efficiency, reduce costs, and improve product quality. In the smart home field, they enable users to remotely control household appliances and adjust the indoor environment, creating a convenient, comfortable, and safe living environment. In industries such as healthcare, transportation, and agriculture, they have also opened a new chapter in intelligent transformation. Due to their profound impact, conducting background research is crucial for comprehensively understanding this product, grasping the development context, clarifying the technological evolution path, providing a theoretical basis for optimizing product performance and expanding application scenarios, and thus promoting high-quality intelligent development of various industries with the help of AIoT modules.

[0003] Currently, there are mainly two semi-hole production processes for AIoT module boards: alkaline etching and acid etching. Alkaline etching is to perform semi-hole processing after graphic electroplating and before alkaline etching. Although it can remove some burrs through etching, there are limitations on the outer layer circuit grade, and products with a line width / line pitch ≤ 60 / 60 μm cannot use this process, thus restricting the production of high-density circuit boards. Acid etching is processed in one step in the final forming process. If the burrs in the half-holes cannot be properly handled during forming, it will cause problems such as the falling off of burrs during component insertion, leading to reliability function problems such as short circuits. That is, it is difficult to balance burr control and process reliability.

[0004] Therefore, an optimized semi-hole burr detection solution for acid etching forming of AIoT module boards is desired. Summary of the Invention

[0005] The present application aims at the deficiencies in the prior art and provides a method and system for detecting burrs in half-holes for acid etching forming of AIoT module boards.

[0006] According to one aspect of the present application, there is provided a method for detecting burrs in half-holes for acid etching forming of AIoT module boards, which includes:

[0007] S1: Provide an AIoT module board;

[0008] S2: Use a common milling cutter to perform rough milling and grooving on a predetermined position of the AIoT module board to obtain a rough milling and grooving area;

[0009] S3: Use a reverse milling cutter to mill the rough milling and grooving area along the left compensation route;

[0010] S4: After performing S3, use a conventional milling cutter to mill the rough fishing slotting area along the right compensation path;

[0011] S5: After performing S4, use a reverse milling cutter to mill the rough fishing slotting area along the left compensation path to obtain a semi-hole with a depression formed at a predetermined position on the AIOT module board;

[0012] S6: Perform burr detection on the semi-hole to obtain a burr detection result, including: performing semantic space enhancement analysis based on the edge features of the detection-standard image on the semi-hole to obtain the burr detection result.

[0013] According to another aspect of the present application, there is provided a semi-hole burr detection system for acid etching forming of an AIOT module board, which includes:

[0014] AIOT module board providing module, used to provide an AIOT module board;

[0015] Rough fishing slotting module, used to perform rough fishing slotting on a predetermined position of the AIOT module board using a conventional milling cutter to obtain a rough fishing slotting area;

[0016] Left compensation reverse milling module, used to mill the rough fishing slotting area along the left compensation path using a reverse milling cutter;

[0017] Right compensation conventional milling module, used to mill the rough fishing slotting area along the right compensation path using a conventional milling cutter after performing the left compensation reverse milling module;

[0018] Semi-hole forming module, used to mill the rough fishing slotting area along the left compensation path using a reverse milling cutter after performing the right compensation conventional milling module to obtain a semi-hole with a depression formed at a predetermined position on the AIOT module board;

[0019] Semi-hole burr detection module, used to perform burr detection on the semi-hole to obtain a burr detection result, wherein the semi-hole burr detection module is used to: perform semantic space enhancement analysis based on the edge features of the detection-standard image on the semi-hole to obtain the burr detection result.

[0020] Due to the adoption of the above technical solutions, the present application has significant technical effects:

[0021] The semi-hole burr detection method and system for AIoT module board acid etching forming provided by this application first use a common milling cutter to rough mill and slot at a predetermined position on the AIoT module board to form a rough milling and slotting area, then use a reverse milling cutter to mill the rough milling and slotting area along the left compensation route, then use a common milling cutter to mill the rough milling and slotting area along the right compensation route, then use the reverse milling cutter again to mill the rough milling and slotting area along the left compensation route to obtain a sunken semi-hole at the predetermined position, and finally perform semantic space enhancement analysis based on the edge features of the detection-standard image on the semi-hole to obtain the burr detection result. In this way, the burr problem generated during the acid etching forming of the AIoT module board can be effectively controlled, which is conducive to the improvement of process reliability. Description of the Drawings

[0022] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0023] Figure 1 It is a flowchart of the semi-hole burr detection method for AIoT module board acid etching forming according to an embodiment of the present application. Figure 2 It is a flowchart of step S6 in the semi-hole burr detection method for AIoT module board acid etching forming according to an embodiment of the present application.

[0024] Figure 3 It is a flowchart of step S65 in the semi-hole burr detection method for AIoT module board acid etching forming according to an embodiment of the present application.

[0025] Figure 4 It is a flowchart of step S66 in the semi-hole burr detection method for AIoT module board acid etching forming according to an embodiment of the present application.

[0026] Figure 5 It is a flowchart of step S663 in the semi-hole burr detection method for AIoT module board acid etching forming according to an embodiment of the present application.

[0027] Figure 6 It is a system block diagram of the semi-hole burr detection system for AIoT module board acid etching forming according to an embodiment of the present application. Detailed Embodiments

[0028] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0029] In the context of the rapid development of today's technology, AIoT module products have promoted the intelligent transformation in fields such as industry, smart home, healthcare, transportation, and agriculture by integrating the Internet of Things and artificial intelligence. In industry, it helps traditional manufacturing industries move towards intelligent manufacturing, improving efficiency and reducing costs; in smart home, it realizes remote control and constructs a comfortable and safe living environment. Conducting background research is of great significance for comprehensively understanding AIoT modules, sorting out the technology evolution path, optimizing performance, and expanding application scenarios, thereby promoting the high-quality intelligent development of the industry. Currently, the semi-hole production process of AIoT module boards mainly uses two methods: alkaline etching and acid etching, but both have technical limitations: alkaline etching is limited in the production of high-density circuit boards, while acid etching is difficult to balance between burr control and process reliability, restricting the further development of the technology.

[0030] Based on this, the present application provides a semi-hole burr detection method for acid etching forming of AIOT module boards. Figure 1 The flowchart of the semi-hole burr detection method for acid etching forming of AIOT module boards according to an embodiment of the present application. As Figure 1 shown, the semi-hole burr detection method for acid etching forming of AIOT module boards according to an embodiment of the present application includes: S1, providing an AIOT module board; S2, using a common milling cutter to perform rough milling and grooving on a predetermined position of the AIOT module board to obtain a rough milling and grooving area; S3, using a reverse milling cutter to mill the rough milling and grooving area along the left compensation route; S4, after performing S3, using a common milling cutter to mill the rough milling and grooving area along the right compensation route; S5, after performing S4, using a reverse milling cutter to mill the rough milling and grooving area along the left compensation route to obtain a semi-hole formed with a depression at a predetermined position of the AIOT module board; S6, performing burr detection on the semi-hole to obtain a burr detection result.

[0031] In step S1, an AIOT module board is provided. It should be understood that the AIOT module board is the physical basis for a series of subsequent processing operations (such as rough milling and grooving, milling, etc.). Different types and specifications of AIOT module boards may have differences in semi-hole processing requirements. Providing a specific AIOT module board can select appropriate processing parameters according to its own design and performance requirements (such as selecting a milling cutter according to the diameter and spacing of the semi-hole to be processed) for subsequent targeted semi-hole production and burr detection, which helps to ensure that the final product meets the quality standards and usage requirements.

[0032] In step S2, a common milling cutter is used to rough mill and slot a predetermined position on the AIOT module board to obtain a rough milling and slotting area. It should be understood that by using a common milling cutter to rough mill and slot a predetermined position on the AIOT module board, a large amount of redundant material can be quickly removed, reducing the machining amount during subsequent fine milling. Moreover, using a common milling cutter in the slotting stage can reduce the possibility of damaging the half hole during slotting, effectively protecting the key parts of the half hole, which provides good basic conditions for subsequent finish machining.

[0033] In step S3, a reverse milling cutter is used to mill the rough milling and slotting area along the left compensation route. Specifically, in the embodiment of the present application, step S3 includes: using a reverse milling cutter to perform milling from the outside of the hole to the inside of the hole on the rough milling and slotting area along the left compensation route to remove the burrs outside the hole. It should be understood that the design of the reverse milling cutter enables it to effectively press down and cut off the burrs during the cutting process, making it more suitable for removing edge burrs compared to a common milling cutter. By using a reverse milling cutter to perform milling from the outside of the hole to the inside of the hole on the rough milling and slotting area along the left compensation route, it can ensure that the burrs outside the hole are removed and will not be redeposited around the hole opening, thus achieving efficient burr removal. It should be noted that the selection of the diameter of the reverse milling cutter needs to be determined according to the diameter of the half hole to be processed and the distance between the two half holes from edge to edge. For example, for products with a half hole diameter ≥ 0.4 mm and a half hole spacing ≥ 0.4 mm, it is recommended to use a reverse milling cutter with a diameter of 0.8 mm; while for products with a half hole diameter ≥ 0.4 mm and a half hole spacing of 0.35 mm, it is recommended to use a reverse milling cutter with a diameter of 0.7 mm.

[0034] In step S4, after performing S3, a common milling cutter is used to mill the rough milling and slotting area along the right compensation route. It should be understood that after the reverse milling cutter removes the burrs outside the hole, the common milling cutter, from different cutting characteristics and directions, mills along the right compensation route, which can clean the burrs outside the hole that the reverse milling cutter may have missed, and can also perform a more comprehensive trimming on the outside edge of the hole, further improving the smoothness of the outside edge of the half hole.

[0035] In step S5, after performing S4, a reverse milling cutter is used to mill the rough fishing slot area along the left compensation route to obtain a semi-hole recessed at a predetermined position on the AIOT module board. Specifically, in the embodiment of the present application, step S5 includes: using a reverse milling cutter to mill the rough fishing slot area from the inside of the hole to the outside along the left compensation route to remove the burrs on the inner side of the hole, and obtaining a semi-hole recessed at a predetermined position on the AIOT module board. It should be understood that by milling from the inside to the outside of the hole, the burrs generated on the inner side of the hole can be removed specifically. This processing direction from the inside to the outside helps to ensure that the burrs are completely removed, while avoiding leaving any residues or re-accumulating burrs inside the hole, thus ensuring the smoothness of the inner edge of the hole. Moreover, using a reverse milling cutter to mill along the left compensation route can perform final fine processing and trimming on the entire semi-hole, making the size and shape of the semi-hole more accurately meet the design requirements. This can lay a good foundation for the assembly and use of the AIOT module board, ensuring the overall performance and quality of the AIoT module board.

[0036] In step S6, burr detection is performed on the semi-hole to obtain a burr detection result. Specifically, in the embodiment of the present application, step S6 includes: performing semantic space enhancement analysis based on the edge features of the detection-standard image on the semi-hole to obtain a burr detection result. It should be understood that in the acid etching forming process, the semi-hole processing is in the final process. If the milling parameters (such as tool wear, feed speed) are not properly controlled, metal burrs are likely to remain on the edge of the hole. Burrs are essentially small protrusions caused by incomplete cutting or plastic deformation of materials during the processing, which are common defects in the manufacturing process. If the burrs fall off, it may cause a short circuit, affecting the electrical performance of the module board and reducing the reliability and stability of the product. At the same time, the existence of burrs may cause the module board not to meet the relevant standards, affecting the market competitiveness of the product and the reputation of the enterprise. By detecting the burrs on the semi-hole. Therefore, by detecting the burrs on the semi-hole, module boards that do not meet the quality requirements can be discovered in time, avoiding unqualified products from flowing into the market. At the same time, it enhances the customer's trust in the product. However, due to the detection of burrs on the semi-hole, because the metal burrs are small in size, complex in shape and highly similar to the normal edge features, traditional manual visual inspection or single image comparison methods have problems such as low efficiency and high missed detection rate. At the same time, existing algorithms are difficult to accurately extract the abnormal difference features (such as local mutations of small burrs) on the edge of the semi-hole under the interference of complex backgrounds, resulting in an increase in the false judgment rate and insufficient reliability of the detection results, and unable to meet the strict requirements of high-precision industrial quality inspection for defect sensitivity and robustness.

[0037] Based on this, the technical concept of this application is to use artificial intelligence-based image processing and feature extraction methods. First, the AOI device automatically acquires the module board image and accurately locates the half-hole area. Combining with the standard template pre-stored in the background, the twin technology is used to extract the edge feature codes of the actual processed image and the template. Through differential feature calculation and spatial-semantic joint enhancement, the local mutation of micro burrs is explicitly modeled, and finally, the robust detection of micro burrs in complex process scenarios is realized. This method amplifies the subtle differences between burrs and normal edges through feature-level enhancement, overcomes the problem of insufficient sensitivity of traditional algorithms to micro defects, significantly improves the detection accuracy and robustness, and at the same time realizes the high-reliability and fully automated determination of half-hole burr defects in complex industrial scenarios, ensuring the electrical performance and product quality of the module board.

[0038] Specifically, Figure 2 FIG. is a flowchart of step S6 in the method for detecting half-hole burrs in the acid etching forming of the AIOT module board according to an embodiment of the present application. As Figure 2 shown, step S6 includes: S61, transporting the AIOT module board to the AOI detection station; S62, scanning the AIOT module board by the AOI device to obtain the AIOT module board image; S63, identifying the region of interest in the AIOT module board image to obtain the half-hole ROI image; S64, extracting the half-hole standard template image from the background database; S65, performing differential calculation based on the image edge features of the half-hole ROI image and the half-hole standard template image to obtain the half-hole edge difference feature; S66, performing spatial-semantic joint edge difference feature enhancement on the half-hole edge difference feature to obtain the half-hole edge difference enhanced feature; S67, based on the half-hole edge difference enhanced feature, obtaining the burr detection result for indicating whether there are burr defects on the AIOT module board.

[0039] In step S61, the AIOT module board is transported to the AOI detection station. It should be understood that the half-hole area of the AIOT module board is extremely delicate, and the micro burrs may be only dozens of micrometers or even smaller. These subtle defects are almost impossible to accurately identify by the human eye. The AOI device at the AOI detection station is equipped with a high-resolution image acquisition system and advanced image processing algorithms, which can accurately capture these micro features. Transporting the AIOT module board to the AOI detection station can clearly distinguish the pixel-level differences between burrs and normal edges using the AOI device at the detection station when detecting the subtle burrs on the half-hole edge, thereby detecting those micro defects that are easily missed by manual detection, and effectively preventing the outflow of unqualified products.

[0040] In step S62, the AIOT module board is scanned by an AOI device to obtain an AIOT module board image. Correspondingly, considering that the AIOT module board has a complex and delicate structure, including numerous tiny features such as lines, components, and half-holes, it is difficult for the human eye to comprehensively and accurately observe these details. The AOI (Automated Optical Inspection) device has high-resolution image acquisition capabilities, can scan the module board in all directions, and obtain a complete and clear image, thus providing a rich data basis for subsequent detection and analysis. Specifically, the AOI device integrates a high-precision optical sensor and a controllable light source system, dynamically adjusts the illumination angle and intensity during the scanning process, effectively suppresses the reflection interference on the metal surface, and enhances the contrast of the burr area. At the same time, the built-in mechanical positioning system of the device ensures that the module board enters the scanning area in a fixed posture, avoiding image distortion caused by position deviation or angle tilt. Through this step, the system can obtain a clear image containing the complete details of the half-hole edge, providing high-quality raw data for subsequent ROI positioning and feature extraction.

[0041] In step S63, the region of interest of the AIOT module board image is identified to obtain a half-hole ROI image. Correspondingly, considering that the detection of half-hole burrs requires high focus on the microscopic features of the hole edge, but the module board image usually contains a large amount of background information irrelevant to the detection (such as pads, lines, identification characters, etc.). These redundant regions not only consume computing resources but also introduce noise interference, making it difficult for the algorithm to effectively extract the subtle abnormalities of the half-hole edge. In addition, the position of the half-hole on the module board may have a slight offset due to production batches or design changes. If the target area is not accurately positioned, it may lead to misalignment of the comparison benchmark and cause misjudgment. Based on this, the present application identifies the region of interest of the AIOT module board image to obtain a half-hole ROI image. That is to say, by concentrating on the half-hole area, the features of this area can be analyzed and processed more accurately. In the entire module board image, interference factors in other areas (such as components of different colors, complex circuit layouts, etc.) may affect the accurate judgment of the half-hole. After extracting the half-hole ROI image, these interference factors can be avoided, and the details of the half-hole can be observed more clearly, thereby improving the accuracy of operations such as burr detection. For example, in the detection of burrs on the half-hole edge, it is easier to find subtle burr features after removing the interference of other areas. In a specific embodiment of the present application, the Faster R-CNN network model is selected to process the AIOT module board image to obtain a half-hole ROI image. The following is a detailed description of a specific implementation process of "identifying the region of interest of the AIOT module board image to obtain a half-hole ROI image":

[0042] Data preparation is the foundation of the entire implementation process. First, a large number of images containing the half-holes of the AIoT module board need to be collected, and special attention should be paid to the diversity of these images. Because in actual industrial production scenarios, the AIoT module board may face different lighting conditions, such as direct strong light, low-light environments, or uneven light reflection; the placement angles of the module boards may also vary, which requires the collected images to contain samples at various angles; at the same time, the complexity of the background cannot be ignored, and there may be other electronic components, circuits, or stains in the background as interfering factors. By collecting rich and diverse images, the subsequent trained model can have stronger generalization ability to adapt to various complex actual situations. After collecting the images, a special annotation tool, such as LabelImg, is used to annotate the half-holes in the images. The annotation information mainly includes the bounding box of the half-hole, that is, the coordinates of the upper left corner and the lower right corner of the rectangular box, and the class label, which is uniformly marked as half-hole here. After annotation, in order to better train and evaluate the model, the annotated data needs to be divided into a training set, a validation set, and a test set, generally in a ratio of 7:2:1.

[0043] Model training is the core step to achieve accurate recognition. The Faster R-CNN model consists of two main parts: the Region Proposal Network (RPN) and the Fast R-CNN detector. The RPN is responsible for generating candidate regions that may contain half-holes, while the Fast R-CNN detector performs precise classification and bounding box regression on these candidate regions. When building the model, deep learning frameworks such as PyTorch or TensorFlow can be used, which provide rich tools and functions, greatly facilitating the construction and training of the model. To improve training efficiency and model performance, a pre-trained convolutional neural network on a large-scale image dataset (such as ImageNet), like ResNet, is usually used as the feature extractor of Faster R-CNN. During the training process, the prepared training set data is input into the model, and the model's parameters are continuously updated by minimizing the classification loss and the bounding box regression loss. At the same time, the validation set is used to monitor the model's performance in real time to avoid overfitting of the model and ensure that the model can accurately learn the features of the half-holes.

[0044] When processing the actual AIOT module board image, image preprocessing is an essential step. First, use an image processing library, such as OpenCV, to read the image to be detected. Since the size of the original image may not meet the input requirements of the model, it is necessary to scale the image to the size that the model can process. During the scaling process, the scaling ratio needs to be recorded because the detection results need to be mapped back to the original image size later to ensure the accuracy of the results. After scaling, it is also necessary to normalize the pixel values of the image, scaling the pixel values to the range of [0,1] or [-1,1], which can accelerate the training and inference speed of the model.

[0045] The Region Proposal Network (RPN) processing is one of the key steps in identifying semi-holes. The preprocessed image is input into a pre-trained convolutional neural network to extract the feature map of the image. The RPN generates a series of anchor boxes with different scales and aspect ratios on the feature map, and these anchor boxes are the candidate regions that may contain semi-holes. Then, the RPN classifies each anchor box to determine whether it contains a semi-hole, and at the same time performs bounding box regression to adjust the position and size of the anchor box to enclose the semi-hole more accurately. Finally, according to the classification scores and the Non-Maximum Suppression (NMS) algorithm, the candidate regions with higher scores and non-overlapping with each other are selected, and these candidate regions will be used as the basis for subsequent processing.

[0046] The FastR-CNN detector will further process the candidate regions generated by the RPN. Extract the features corresponding to the candidate regions from the feature map, and then classify these candidate regions to determine whether they are semi-holes. At the same time, perform bounding box regression again to further optimize the position and size of the candidate regions. After this step, the position and range of the semi-holes can be judged more accurately. Then, according to the classification scores and the NMS algorithm, the detection results with higher scores and non-overlapping with each other are selected, and these are the finally determined semi-hole detection results.

[0047] Finally, extract the semi-hole ROI image according to the detection results. First, map the bounding box coordinates of the detection results from the scaled image back to the original image size to ensure the accuracy of the coordinates. Then, use the image processing library OpenCV to crop the semi-hole region from the original image according to the mapped bounding box coordinates, and thus obtain the semi-hole ROI image.

[0048] In step S64, a semi-hole standard template image is extracted from the background database. It should be understood that the extracted semi-hole standard template image contains geometric shape information, edge detail information, texture information, etc. of the semi-hole in an ideal state. When detecting the semi-hole state of the AIOT module board, comparing the actually obtained semi-hole ROI image with the standard template image in terms of feature dimensions can more accurately discover the differences between the actual semi-hole and the ideal state, so as to accurately judge whether there are burrs on the actual semi-hole, thereby improving the reliability of the burr detection result.

[0049] In step S65, a difference calculation based on the image edge features is performed on the semi-hole ROI image and the semi-hole standard template image to obtain the semi-hole edge difference feature. Specifically, Figure 3 FIG. is a flowchart of step S65 in the semi-hole burr detection method for AIOT module board acid etching forming according to an embodiment of the present application. As Figure 3 shown, step S65 includes: S651, inputting the semi-hole ROI image and the semi-hole standard template image into a semi-hole edge feature siamese detection network to obtain a semi-hole edge feature encoded feature map and a semi-hole edge template feature encoded feature map; S652, calculating a semi-hole edge difference feature map between the semi-hole edge feature encoded feature map and the semi-hole edge template feature encoded feature map as the semi-hole edge difference feature.

[0050] In step S651, the semi-hole ROI image and the semi-hole standard template image are input into the semi-hole edge feature siamese detection network to obtain the semi-hole edge feature encoded feature map and the semi-hole edge template feature encoded feature map. Correspondingly, considering that the semi-hole ROI image and the semi-hole standard template image contain a large amount of pixel information, but not all information is relevant to the semi-hole edge feature. Therefore, in order to extract key information closely related to the semi-hole edge feature from complex images to more accurately describe the edge characteristics of the semi-hole, such as the shape, continuity, curvature, etc. of the edge, and provide a more effective data basis for subsequent comparative analysis. In this application, the semi-hole ROI image and the semi-hole standard template image are input into the semi-hole edge feature siamese detection network to obtain the semi-hole edge feature encoded feature map and the semi-hole edge template feature encoded feature map. It should be understood that through the dual-branch structure of the siamese network, the deep feature expressions of the actual processed image and the standard template can be learned synchronously. Specifically, the network adopts a dual-channel design with shared weights, and performs multi-scale convolution and attention mechanism processing on the ROI image and the template image respectively to extract the encoded feature map containing edge geometric shapes, texture distributions, and local mutation information. Among them, the ROI branch focuses on the real-time features of the currently processed semi-hole (such as the micron-level protrusions of burrs), and the template branch represents the reference pattern of the ideal semi-hole edge (such as a smooth contour, uniform transition). Through the feature alignment ability of the siamese structure, the network can adaptively match the semantic spaces of the two images, weaken the influence of background noise and process fluctuations, and at the same time amplify the subtle differences between burrs and normal edges. For example, when the edges of a batch of semi-holes are slightly thickened due to tool wear, the network can distinguish the reasonable changes within the process tolerance range from the abnormal mutations of real burrs by comparing the reference features of the template.

[0051] In step S652, calculate the semi-hole edge difference feature map between the semi-hole edge feature encoding feature map and the semi-hole edge template feature encoding feature map as the semi-hole edge difference feature. It should be understood that the semi-hole edge feature encoding feature map and the semi-hole edge template feature encoding feature map respectively represent the edge features of the actual semi-hole and the standard semi-hole. By directly observing these two feature maps, it is difficult to intuitively see the differences between them. Therefore, in the technical solution of this application, calculate the semi-hole edge difference feature map between the semi-hole edge feature encoding feature map and the semi-hole edge template feature encoding feature map as the semi-hole edge difference feature. Specifically, the feature encoding extracted by the Siamese network is not simply a set of pixel values, but an abstract semantic representation learned through multiple layers of convolution and attention mechanisms - for example, the encoding feature map may contain deep information such as edge curvature, texture continuity, and local gradient mutations. In this feature space, the morphological fluctuations of normal edges (such as slight offsets allowed by process parameters) are represented as a low-dimensional continuous distribution, while the abnormal protrusions of burrs are mapped as discrete outliers in the high-dimensional feature space. By calculating the per-channel differences (such as cosine distance or Euclidean distance) between the two feature maps, the deviation between the actually processed semi-hole and the standard template at the semantic level can be quantified, while suppressing the random differences caused by noise. For example, when a certain area shows significant differences in the pixel-level image due to reflection, it may be encoded as a low response value in the feature space and thus automatically filtered by the difference calculation; while the real burrs, due to their unique geometric mutation patterns, will have their difference features significantly amplified.

[0052] In step S66, perform spatial-semantic joint edge difference feature enhancement on the semi-hole edge difference feature to obtain the semi-hole edge difference enhanced feature. Specifically, Figure 4 FIG. is a flowchart of step S66 in the semi-hole burr detection method for AIOT module board acid etching forming according to an embodiment of the present application. As Figure 4 shown, step S66 includes: S661, extract the channel feature vector at the (i, j) pixel position from the semi-hole edge difference feature map as the semi-hole edge difference channel feature vector to be enhanced; S662, perform n random scans on the semi-hole edge difference feature map to obtain n semi-hole edge difference channel feature vectors as a sparse set of semi-hole edge difference reference feature vectors; S663, based on the sparse set of semi-hole edge difference reference feature vectors, perform spatial modulation and semantic association compensation on the semi-hole edge difference channel feature vector to be enhanced to obtain the semi-hole edge difference enhanced feature map as the semi-hole edge difference enhanced feature.

[0053] It should be understood that although the semi-hole edge difference feature map has extracted the deep differences between the actual processed image and the standard template through the Siamese network, the directly output difference signal still faces multiple interferences: on the one hand, metal reflection, surface texture noise, and normal edge offsets caused by process fluctuations in complex industrial scenarios may lead to a large number of non-defect-related signals being mixed in the difference feature map; on the other hand, the local mutation features of micron-scale burrs are limited in differentiating from noise in the traditional feature space due to their small size and diverse shapes, and it is difficult to achieve high-confidence defect determination based solely on the initial difference map. For example, a slight serration of the normal edge caused by tool vibration in a certain area may be misjudged as a burr, while the true burr may be weakened or masked by the difference map due to surrounding background interference, resulting in missed detection. Therefore, in the technical solution of this application, spatial-semantic joint edge difference feature enhancement is performed on the semi-hole edge difference feature map to obtain the semi-hole edge difference enhanced feature map as the semi-hole edge difference enhanced feature. That is, this step aims to reconstruct the expression space of the difference features through a spatial-semantic double-layer modulation mechanism based on the Poincaré distance and implicit semantic association, and achieve the collaborative optimization of noise suppression and defect signal enhancement. Specifically, the spatial level can consider information such as the position and shape of the difference features in the image, and the semantic level combines the understanding of the image content, such as the normal shape of the semi-hole edge and the possible defect shapes. Through this joint enhancement, the originally unobvious subtle differences become more prominent, facilitating subsequent analysis and judgment.

[0054] Specifically, first, the channel feature vector at the (i, j) pixel position is extracted from the semi-hole edge difference feature map as the semi-hole edge difference channel feature vector to be enhanced. The above process can be expressed by the formula:

[0055] F ∈ R H×W×C

[0056] v tbs = F(i, j, :) ∈ R C

[0057] where F is the semi-hole edge difference feature map, R is the set of real numbers, H and W are the height and width of each feature matrix of F along the channel dimension, C is the number of channels of F, F(i, j, :) is the channel feature vector at the (i, j) pixel position in F, and v tbs is the semi-hole edge difference channel feature vector to be enhanced, R cIt is a sparse set of reference feature vectors of the semi-hole edge difference. It should be understood that the semi-hole edge difference feature map contains a large amount of information, and a multi-dimensional channel feature vector corresponding to each pixel position represents a specific feature representation of that local area. By extracting the channel feature vector at the (i,j) pixel position as the semi-hole edge difference channel feature vector to be enhanced, it is essentially an example of the feature enhancement process of the channel feature vector at the (i,j) pixel position to illustrate the feature enhancement process of the semi-hole edge difference feature map.

[0058] Next, the semi-hole edge difference feature map is randomly scanned n times to obtain n semi-hole edge difference channel feature vectors as a sparse set of reference feature vectors of the semi-hole edge difference. The above process can be expressed by the formula:

[0059] R c ={r1,r2,...,r i ,...,r n}

[0060] where R c is a sparse set of reference feature vectors of the semi-hole edge difference, r1, r2, r i and r n are the 1st, 2nd, i-th and n-th semi-hole edge difference reference feature vectors in the sparse set of reference feature vectors of the semi-hole edge difference respectively, and r i ∈R C .

[0061] It should be understood that traditional fixed sampling is difficult to comprehensively and accurately extract key information, while random scanning can introduce spatial diversity by randomly sampling channel feature vectors at different positions of the semi-hole edge difference feature map multiple times, avoiding being limited to some local areas of the feature map due to fixed sampling and thus falling into the dilemma of local overfitting. Specifically, on the one hand, the sparse set of reference feature vectors of the semi-hole edge difference formed by performing n random scans essentially constitutes a kind of feature memory bank. Each reference feature vector of the semi-hole edge difference in the set represents a local semantic information in the original feature map, and these information are rich and diverse, which can provide sufficient reference basis for the subsequent determination of the existence of burrs. On the other hand, sparsity makes it easier for the model to capture more representative feature patterns in the feature space, effectively avoiding being interfered by redundant or noise features. For example, when facing the coexistence of slight serrations on the normal edge caused by tool vibration and real burrs, this sparse set can, with rich, diverse and representative feature patterns, enable the model to accurately distinguish normal edge changes from burr defects, which is conducive to improving the accuracy and reliability of semi-hole burr detection.

[0062] Specifically, Figure 5It is a flowchart of step S663 in the method for detecting burrs on half - holes in the acid etching forming of the AIOT module board according to an embodiment of the present application. As Figure 5 shown, step S663 includes: S6631, calculating the Poincaré distance between the feature vector of the edge difference channel of the half - hole to be enhanced and each reference feature vector of the edge difference of the half - hole in the sparse set of reference feature vectors of the edge difference of the half - hole to obtain the spatial modulation matrix of the edge difference of the half - hole; S6632, calculating the implicit semantic association between the feature vector of the edge difference channel of the half - hole to be enhanced and each reference feature vector of the edge difference of the half - hole in the sparse set of reference feature vectors of the edge difference of the half - hole to obtain a set of semantic association coding matrices of the edge difference of the half - hole; S6633, using each semantic association coding matrix in the set of semantic association coding matrices of the edge difference of the half - hole as the primary mask modulation unit and using the spatial modulation matrix of the edge difference of the half - hole as the secondary mask modulation unit, and performing explicit modeling modulation on the information compensation coding vector of the edge difference between each reference feature vector of the edge difference of the half - hole in the sparse set of reference feature vectors of the edge difference of the half - hole and the feature vector of the edge difference channel of the half - hole to be enhanced to obtain the implicit coding vector of the enhanced component of the edge difference of the half - hole; S6634, fusing the implicit coding vector of the enhanced component of the edge difference of the half - hole and the feature vector of the edge difference channel of the half - hole to be enhanced to obtain the enhanced feature vector of the edge difference channel of the half - hole, where the enhanced feature vector of the edge difference channel of the half - hole is the channel feature vector at the (i, j) pixel position of the edge difference enhanced feature map.

[0063] Then, calculate the Poincaré distance between the feature vector of the edge difference channel of the half - hole to be enhanced and each reference feature vector of the edge difference of the half - hole in the sparse set of reference feature vectors of the edge difference of the half - hole to obtain the spatial modulation matrix of the edge difference of the half - hole. The above process can be expressed by the formula:

[0064]

[0065] where, ‖·‖ 2 is the square of calculating the Euclidean norm, arccosh is the inverse hyperbolic cosine function, M tbsii is the Poincaré distance between v tbs and r i , M tbs- and M tbsin are the respective eigenvalues in the spatial modulation matrix of the edge difference of the half - hole, and M s is the spatial modulation matrix of the edge difference of the half - hole.

[0066] It should be understood that the features of the half-hole edge difference channel to be enhanced have hierarchical distribution characteristics. The traditional Euclidean distance is difficult to accurately capture the similarities and differences between these features, while the Poincare distance is more suitable for measuring such hierarchical data, and can more accurately describe the relationship between the feature vector of the half-hole edge difference channel to be enhanced and the reference feature vector of the half-hole edge difference. Specifically, the half-hole edge difference spatial modulation matrix calculated by the Poincare distance can better model the hierarchical relationship in the feature space, which makes it possible to enhance the consistency between the feature vector of the half-hole edge difference channel to be enhanced and the reference feature vector of the half-hole edge difference in the feature enhancement process, thereby improving the distinguishing ability of the feature representation. For example, when distinguishing the normal morphology and burr defects of the half-hole edge, the feature differences brought by the burr can be more accurately highlighted, thereby improving the accuracy of detection. In addition, the generated half-hole edge difference spatial modulation matrix can guide the subsequent feature enhancement process to focus on information from areas with strong spatial correlation. For the half-hole edge difference reference feature vector corresponding to the high Poincare distance value, its difference or complementarity in the nonlinear space can provide more information supplement for feature enhancement and avoid being interfered by redundant or noise features. This can more effectively extract and enhance the features related to burr defects at different positions in the feature map, thereby reducing the occurrence of misjudgment and missed detection. Next, the implicit semantic association between the half-hole edge difference channel feature vector to be enhanced and each half-hole edge difference reference feature vector in the sparse set of half-hole edge difference reference feature vectors is calculated to obtain a set of half-hole edge difference semantic association encoding matrices. The above process can be expressed by the formula:

[0067]

[0068]

[0069] in, is the matrix-matrix multiplication, W i For r i The corresponding weight matrix, T is the transpose operation, d is W i and r i The length of the vector after multiplication, softmax is the softmax function, Yes tbs and r i The semi-hole edge difference semantic association coding matrix between them, that is, the i-th semi-hole edge difference semantic association coding matrix in the set of semi-hole edge difference semantic association coding matrices, M semantic is the set of semi-hole edge difference semantic association encoding matrices, and They are respectively the first, second and nth half-hole edge difference semantic association coding matrices in the set of half-hole edge difference semantic association coding matrices.

[0070] It should be understood that due to the complex interference faced by the half-hole edge difference channel feature to be enhanced, it is difficult to accurately identify defect features such as burrs by relying solely on the surface similarity at the pixel level. However, implicit semantic association goes beyond this surface similarity and is committed to mining the high-level concepts or patterns contained in the features. Specifically, when there are micron-level burrs on the edge of the half-hole, the changes at the pixel level may not be obvious, but through implicit semantic association, the changes in the features brought by the burrs at the semantic level can be captured, such as the differences in high-level concepts such as morphology and structure from the normal half-hole edge, so as to more accurately detect the burr defects. The values ​​in the generated half-hole edge difference semantic association encoding matrix reflect the degree of semantic similarity between the feature vector to be enhanced and the reference vector. Through this half-hole edge difference semantic association encoding matrix, the model can clarify the importance of different half-hole edge difference reference feature vectors to the half-hole edge difference channel feature vector to be enhanced at the semantic level in the analysis. For the half-hole edge difference reference feature vector with a high degree of semantic similarity to the half-hole edge difference channel feature vector to be enhanced, more attention and weight can be given in the subsequent feature enhancement process to improve the feature's expressiveness and discrimination ability.

[0071] More specifically, in the embodiment of the present application, step S6633 includes: calculating the position difference vector between the half-hole edge difference channel feature vector to be enhanced and each half-hole edge difference reference feature vector to obtain a set of half-hole edge difference information compensation coding vectors; based on the primary mask modulation unit and the secondary mask modulation unit, performing strong and weak coupling of double-field interaction on each half-hole edge difference information compensation coding vector in the set of half-hole edge difference information compensation coding vectors to obtain a set of optimized half-hole edge difference information compensation coding vectors; based on the primary mask modulation unit and the secondary mask modulation unit, performing explicit modeling modulation on the set of optimized half-hole edge difference information compensation coding vectors to obtain an implicit coding vector of the half-hole edge difference enhancement component. The above overall process can be expressed by the formula:

[0072] x i =r i -v tbs

[0073]

[0074]

[0075] Among them, x i Yes tbs and r i The half-hole edge difference information between the two is the compensation encoding vector, x i ′ Yes tbs and r iSemantic enhancement coding vector for compensating semi-hole edge difference information therebetween, x i ″ is v tbs and r i Spatial enhancement coding vector for compensating semi-hole edge difference information therebetween, n is the number of vectors in the sparse set of semi-hole edge difference reference feature vectors, Δv tbs is the implicit coding vector of the semi-hole edge difference enhancement component.

[0076] Specifically, the semi-hole edge difference information compensation coding vector to be modulated is regarded as a modulation field density distribution with dynamic propagation characteristics, and its physical connotation lies in characterizing the intensity distribution law of feature compensation at different spatial positions through the field density gradient. Under this framework, the set of semantic association coding matrices as the first-level mask modulation unit can be analogized to the gauge field, and the semantic symmetry transformation in the feature space is constrained through its tensor structure to ensure that the feature enhancement process still maintains the gauge invariance with the core semantics of the burr defect under interference conditions such as tool vibration and metal reflection; while the spatial modulation matrix corresponding to the second-level mask modulation unit serves as the covariant field, and its non-Euclidean geometric properties can adaptively adjust the local curvature of the feature space. When dealing with the normal edge deformation caused by process fluctuations, the geometric invariance of the feature compensation process is maintained through the covariant derivative. In specific implementation, first, the Riemannian metric calibration of the feature space is performed based on the gauge field, and a hierarchical semantic association network is established using the Poincaré distance, so that the information compensation coding vector remains invariant in the semantic core under gauge transformation; then, through the affine connection structure of the covariant field, a covariant differential equation of the spatial modulation degree is established in the local coordinate system of the feature manifold to achieve geometric compensation for spatial interferences such as position offset and morphological distortion. This dual-field coupling mechanism essentially constructs an interaction model of the gauge-covariant field: the gauge field maintains global semantic consistency through the fiber bundle structure, and the covariant field regulates local geometric deformation through the Levi-Civita connection. The two achieve strong and weak coupling through the ordinary partial derivative term in the field equation - the strong coupling region corresponds to the microscopic mutation characteristics of the burr defect, and its field density gradient is significantly enhanced, forming topological defects on the high-dimensional feature manifold; the weak coupling region corresponds to the smooth transition of the normal edge, and the field interaction tends to be evenly distributed. This field theory modeling method based on differential geometry enables the double-layer mask mechanism to break through the limitations of traditional linear modulation, establish a more accurate information compensation correlation in the non-linear feature space, and ultimately, through the co-evolution of the gauge-covariant field, while maintaining the stability of the semantic core, dynamically adapt to the spatial geometric distortion in complex industrial scenarios, thereby significantly improving the effectiveness and robustness of micron-level burr feature coding.

[0077] It should be understood that each matrix in the set of semantic association coding matrices for the semi-hole edge difference is used as a first-level mask modulation unit. Based on knowledge such as the normal shape and possible defect shapes of the semi-hole edge, the model can perform preliminary semantic filtering on the feature vectors, clarify which are the key semantic information related to the semi-hole edge defects, and exclude irrelevant semantic interferences such as metal reflection and surface texture noise in complex industrial scenarios, thus focusing on the defective semantic part that really needs attention; using the semi-hole edge difference spatial modulation matrix as the second-level mask modulation unit can screen the feature vectors from the spatial levels such as the position and shape of the semi-hole edge, ensuring that spatial interference factors such as normal edge offsets caused by process fluctuations do not affect the judgment of burr defects, and effectively balancing the contributions of global semantic information and local spatial information. In this process, explicit modeling modulation is performed on the optimized semi-hole edge difference information compensation coding vector between each semi-hole edge difference reference feature vector and the semi-hole edge difference channel feature vector to be enhanced, so that the model pays more attention to the difference or incremental part of the semi-hole edge difference reference feature vector relative to the semi-hole edge difference channel feature vector to be enhanced during analysis. For example, when a certain area is suspected to have burrs, the difference information in the reference vector regarding the normal semi-hole edge and the current area to be detected becomes the key focus. In this way, the enhancement process is more focused on the information part that needs to be supplemented and corrected, and can accurately extract and effectively utilize the key information in the reference vector. The finally obtained semi-hole edge difference enhanced component implicit coding vector, as an intermediate representation, encodes the enhanced components extracted from the semi-hole edge difference reference feature vector. In actual detection scenarios, this implicit coding strategy makes the entire enhancement process more flexible and plastic. It allows the detection model to learn complex non-linear enhancement functions according to different actual situations of the semi-hole, so as to more accurately identify the local mutation characteristics of micron-level burrs, avoiding misjudgment and missed detection caused by factors such as small burr size, diverse shapes, and background interference, and greatly improving the effectiveness and robustness of semi-hole burr detection.

[0078] Finally, fuse the semi-hole edge difference enhanced component implicit coding vector and the semi-hole edge difference channel feature vector to be enhanced to obtain the enhanced semi-hole edge difference channel feature vector, where the enhanced semi-hole edge difference channel feature vector is the channel feature vector at the (i, j) pixel position of the semi-hole edge difference enhanced feature map. The above process can be expressed by the formula:

[0079] vi enhanced = α·Δv tbs + β·v tbs

[0080] where α and β are weighted hyperparameters, and vi enhanced is r iThe enhanced enhanced semi-hole edge difference channel feature vector, i.e., the channel feature vector at the (i, j) pixel position of the hole edge difference enhancement feature map.

[0081] It should be understood that the semi-hole edge difference enhancement component implicit coding vector encodes the key enhancement components extracted from the reference vector, which focus on supplementing and correcting the difference information between the semi-hole edge to be detected and the standard template, especially for the feature enhancement of fine defects such as micron-level burrs. The semi-hole edge difference channel feature vector to be enhanced contains the basic discriminant information of the semi-hole edge, but its expression ability for burr defects is still insufficient under complex background interference. By fusing these two vectors, effective enhancement at the feature level can be achieved. For example, the discriminant information such as the basic shape and position of the semi-hole edge in the original enhanced semi-hole edge difference channel feature vector is retained, which ensures that the key features of the semi-hole itself are not lost during the detection process, making the detection result always based on the actual shape of the semi-hole. At the same time, the incorporated enhancement information can utilize the relevant knowledge of the context environment to more accurately identify the areas where burrs may exist. For example, in the face of interference factors such as metal reflection, surface texture noise, and normal edge offsets caused by process fluctuations in complex industrial scenarios, the enhancement information can help better distinguish true burr defects from these interference signals, improving the sensitivity and anti-interference ability of the feature vector to burr defects. This provides a more reliable and distinguishable feature basis for accurately judging whether there are burr defects on the semi-hole edge in the subsequent process.

[0082] In step S67, based on the enhanced features of the semi-hole edge differences, a burr detection result is obtained to indicate whether there are burr defects on the AIOT module board. Specifically, in the embodiment of the present application, step S67 includes: passing the enhanced feature map of the semi-hole edge differences through a defect detector based on a classifier to obtain the burr detection result, and the burr detection result is used to indicate whether there are burr defects on the AIOT module board. That is, the fabric surface texture-foreground morphology response fusion feature vector obtained by feature enhancement using the semi-hole edge difference feature map is classified to intelligently determine whether there are burr defects. It should be understood that a classifier is a mature machine learning tool. After being trained with a large number of samples, it can learn the relationship between different features and defects. The enhanced feature map of the semi-hole edge differences contains rich feature information related to burr defects. Inputting it into the defect detector based on the classifier can make full use of the classifier's classification and judgment capabilities for features. The classifier can analyze and identify the features in the feature map according to the pre-trained model, so as to accurately determine whether there are burr defects and improve the accuracy and reliability of detection. In this way, accurately detecting these defects can timely screen out unqualified products and prevent defective module boards from being applied to actual products, thereby ensuring the quality and reliability of AIOT products. In a specific embodiment of the present application, passing the enhanced feature map of the semi-hole edge differences through a defect detector based on a classifier to obtain the burr detection result includes: expanding each semi-hole edge difference enhanced feature matrix in the enhanced feature map of the semi-hole edge differences into a one-dimensional feature vector according to a row vector or a column vector and then cascading them to obtain a semi-hole burr detection classification feature vector; using the fully connected layer of the classifier to perform fully connected encoding on the semi-hole burr detection classification feature vector to obtain a semi-hole burr detection encoded classification feature vector; inputting the semi-hole burr detection encoded classification feature vector into the Softmax classification function of the classifier to obtain the burr detection result.

[0083] In summary, step S6 is clearly described. It uses an artificial intelligence-based image processing and feature extraction method. First, the AOI device automatically acquires the module board image and accurately locates the semi-hole area. Combining with the standard template pre-stored in the background, the edge feature codes of the actual processed image and the template are extracted using the siamese technology. Through difference feature calculation and spatial-semantic joint enhancement, the local mutation of micro burrs is explicitly modeled, and finally, the robust detection of micro burrs in complex process scenarios is realized. In this way, by enhancing the feature level to amplify the subtle differences between burrs and normal edges, the problem of insufficient sensitivity of traditional algorithms to micro defects is overcome, the detection accuracy and robustness are significantly improved, and at the same time, the determination of semi-hole burr defects with high reliability and full automation in complex industrial scenarios is realized, ensuring the electrical performance of the module board and the product quality.

[0084] In summary, a method for detecting burrs in half-holes for acid etching forming of an AIoT module board according to an embodiment of the present application is elucidated. First, a rough milling slot is formed in a predetermined position of the AIoT module board with a common milling cutter to form a rough milling slot area. Then, the rough milling slot area is milled along a left compensation route with a reverse milling cutter. Next, the rough milling slot area is milled along a right compensation route with a common milling cutter. Subsequently, the rough milling slot area is milled again along the left compensation route with a reverse milling cutter to obtain a sunken half-hole in the predetermined position. Finally, semantic space enhancement analysis based on the edge features of the detected-standard image is carried out on the half-hole to obtain the burr detection result. In this way, the burr problem generated during the acid etching forming of the AIoT module board can be effectively controlled, which is conducive to the improvement of process reliability.

[0085] Figure 6 FIG. is a system block diagram of a half-hole burr detection system for acid etching forming of an AIoT module board according to an embodiment of the present application. As Figure 6 shown, the half-hole burr detection system 100 for acid etching forming of an AIoT module board according to an embodiment of the present application includes: an AIoT module board providing module 110 for providing an AIoT module board; a rough milling slotting module 120 for using a common milling cutter to perform rough milling slotting on a predetermined position of the AIoT module board to obtain a rough milling slot area; a left compensation reverse milling module 130 for using a reverse milling cutter to mill the rough milling slot area along a left compensation route; a right compensation common milling module 140 for, after executing the left compensation reverse milling module, using a common milling cutter to mill the rough milling slot area along a right compensation route; a half-hole forming module 150 for, after executing the right compensation common milling module, using a reverse milling cutter to mill the rough milling slot area along a left compensation route to obtain a half-hole formed in a sunken manner at a predetermined position of the AIoT module board; and a half-hole burr detection module 160 for detecting burrs in the half-hole to obtain a burr detection result.

[0086] Here, those skilled in the art can understand that the specific functions and operations of each unit and module in the above half-hole burr detection system 100 for acid etching forming of an AIoT module board have been introduced in detail in the description of the above Figures 1 to 5 half-hole burr detection method for acid etching forming of an AIoT module board, and therefore, the repeated description thereof will be omitted.

[0087] In summary, the half-hole burr detection system 100 for the acid etching forming of the AIoT module board based on the embodiments of the present application is elucidated. First, a rough milling slot is formed in a predetermined position of the AIoT module board with a common milling cutter to form a rough milling slot area. Then, the rough milling slot area is milled along the left compensation route with a reverse milling cutter. Next, the rough milling slot area is milled along the right compensation route with a common milling cutter. Subsequently, the rough milling slot area is milled again along the left compensation route with a reverse milling cutter to obtain a sunken half-hole in the predetermined position. Finally, semantic space enhancement analysis based on the edge features of the detection-standard image is carried out on the half-hole to obtain the burr detection result. In this way, the burr problem generated during the acid etching forming of the AIoT module board can be effectively controlled, which is beneficial to the improvement of process reliability.

Claims

1. A method for detecting half-hole burrs in acid etching of AIOT module boards, characterized in that: include: S1: Provide AIOT module board; S2: using a common milling cutter to perform rough grooving on a predetermined position of the AIOT module board to obtain a rough grooving area; S3: milling the rough grooving area along the left compensation route using a reverse milling cutter; S4: after executing S3, the roughing slotting area is milled along the right compensation route using the common milling cutter; S5: after executing S4, using the reverse milling cutter to mill the roughing slot area along the left compensation route to obtain a half hole recessed at a predetermined position of the AIOT module board; S6: performing burr detection on the half hole to obtain a burr detection result, including: performing semantic space enhancement analysis based on detection-standard image edge features on the half hole to obtain the burr detection result.

2. The half-hole burr detection method for acid etching forming of AIOT module board according to claim 1 is characterized in that: The step S3 includes: using the reverse milling cutter to perform milling processing on the rough grooving area from outside the hole to inside the hole along the left compensation route to remove burrs on the outside of the hole.

3. The half-hole burr detection method for acid etching forming of AIOT module board according to claim 1 is characterized in that: The step S5 includes: using the reverse milling cutter to perform milling processing on the rough groove area from inside the hole to outside the hole along the left compensation route to remove burrs inside the hole, thereby obtaining the half hole recessed at a predetermined position of the AIOT module board.

4. The half-hole burr detection method for acid etching molding of AIOT module board according to claim 1 is characterized in that: The step S6 comprises: Transporting the AIOT module board to the AOI inspection station; Scanning the AIOT module board by an AOI device to obtain an AIOT module board image; Performing region of interest recognition on the AIOT module board image to obtain a half-hole ROI image; Extracting half-hole standard template image from the background database; Performing a difference calculation based on image edge features on the half-hole ROI image and the half-hole standard template image to obtain a half-hole edge difference feature; Performing spatial-semantic joint edge difference feature enhancement on the half-hole edge difference feature to obtain a half-hole edge difference enhancement feature; Based on the half-hole edge difference enhancement feature, the burr detection result is obtained to indicate whether the AIOT module board has a burr defect.

5. The half-hole burr detection method for acid etching forming of AIOT module board according to claim 4 is characterized in that: The half-hole ROI image and the half-hole standard template image are subjected to difference calculation based on image edge features to obtain half-hole edge difference features, including: Inputting the half-hole ROI image and the half-hole standard template image into a half-hole edge feature twin detection network to obtain a half-hole edge feature encoding feature map and a half-hole edge template feature encoding feature map; A half-hole edge difference feature map between the half-hole edge feature coding feature map and the half-hole edge template feature coding feature map is calculated as the half-hole edge difference feature.

6. The method for detecting half-hole burrs in the acid etching molding of AIOT module boards according to claim 4, characterized in that: The half-hole edge difference feature is subjected to spatial-semantic joint edge difference feature enhancement to obtain a half-hole edge difference enhancement feature, including: Extracting a channel feature vector at the (i, j)th pixel position from the half-hole edge difference feature map as the half-hole edge difference channel feature vector to be enhanced; Performing n random scans on the half-hole edge difference feature map to obtain n half-hole edge difference channel feature vectors as a sparse set of half-hole edge difference reference feature vectors; Based on the sparse set of the half-hole edge difference reference feature vectors, the half-hole edge difference channel feature vectors to be enhanced are spatially modulated and semantically associated compensated to obtain a half-hole edge difference enhancement feature map as the half-hole edge difference enhancement feature.

7. The half-hole burr detection method for acid etching forming of AIOT module board according to claim 6 is characterized in that: Based on the sparse set of the half-hole edge difference reference feature vectors, spatial modulation and semantic association compensation are performed on the half-hole edge difference channel feature vectors to be enhanced to obtain a half-hole edge difference enhanced feature map, including: Calculating the Poincare distance between the half-hole edge difference channel feature vector to be enhanced and each half-hole edge difference reference feature vector in the sparse set of half-hole edge difference reference feature vectors to obtain a half-hole edge difference spatial modulation matrix; Calculating implicit semantic associations between the half-hole edge difference channel feature vector to be enhanced and each half-hole edge difference reference feature vector in the sparse set of half-hole edge difference reference feature vectors to obtain a set of half-hole edge difference semantic association encoding matrices; Using each half-hole edge difference semantic association coding matrix in the set of the half-hole edge difference semantic association coding matrices as a primary mask modulation unit and using the half-hole edge difference spatial modulation matrix as a secondary mask modulation unit, performing explicit modeling modulation on the half-hole edge difference information compensation coding vector between each half-hole edge difference reference feature vector in the sparse set of the half-hole edge difference reference feature vectors and the half-hole edge difference channel feature vector to be enhanced to obtain an implicit coding vector of the half-hole edge difference enhancement component; The half-hole edge difference enhancement component implicit coding vector and the half-hole edge difference channel feature vector to be enhanced are fused to obtain an enhanced half-hole edge difference channel feature vector, wherein the enhanced half-hole edge difference channel feature vector is the channel feature vector of the (i, j)th pixel position of the half-hole edge difference enhancement feature map.

8. The method for detecting half-hole burrs in the acid etching molding of AIOT module boards according to claim 7, characterized in that: Using each half-hole edge difference semantic association coding matrix in the set of the half-hole edge difference semantic association coding matrices as a primary mask modulation unit and using the half-hole edge difference spatial modulation matrix as a secondary mask modulation unit, performing explicit modeling modulation on the half-hole edge difference information compensation coding vector between each half-hole edge difference reference feature vector in the sparse set of the half-hole edge difference reference feature vector and the half-hole edge difference channel feature vector to be enhanced to obtain an implicit coding vector of the half-hole edge difference enhancement component, including: Calculating the position difference vector between the half-hole edge difference channel feature vector to be enhanced and each half-hole edge difference reference feature vector to obtain a set of half-hole edge difference information compensation coding vectors; Based on the primary mask modulation unit and the secondary mask modulation unit, each half-hole edge difference information compensation coding vector in the set of half-hole edge difference information compensation coding vectors is subjected to strong and weak coupling of double-field interaction to obtain a set of optimized half-hole edge difference information compensation coding vectors; Based on the primary mask modulation unit and the secondary mask modulation unit, explicit modeling modulation is performed on the set of optimized half-aperture edge difference information compensation coding vectors to obtain the half-aperture edge difference enhancement component implicit coding vector.

9. The method for detecting half-hole burrs in the acid etching molding of AIOT module boards according to claim 8, characterized in that: Based on the half-hole edge difference enhancement feature, the burr detection result is obtained to indicate whether the AIOT module board has a burr defect, including: passing the half-hole edge difference enhancement feature map through a classifier-based defect detector to obtain the burr detection result, and the burr detection result is used to indicate whether the AIOT module board has a burr defect.

10. A half-hole burr detection system for acid etching of AIOT module boards, characterized in that: include: The AIOT module board provides a module for providing an AIOT module board; A rough grooving module, used for rough grooving a predetermined position of the AIOT module board using a common milling cutter to obtain a rough grooving area; A left compensation reverse milling module, used for milling the rough grooving area along a left compensation route using a reverse milling cutter; A right compensation common milling module, used for milling the roughing slotting area along the right compensation route using the common milling cutter after executing the left compensation reverse milling module; A half-hole forming module, used for, after executing the right compensation ordinary milling module, using the reverse milling cutter to mill the rough grooving area along the left compensation route to obtain a half-hole concavely formed at a predetermined position of the AIOT module board; A half-hole burr detection module is used to perform burr detection on the half-hole to obtain a burr detection result, wherein the half-hole burr detection module is used to: perform semantic space enhancement analysis on the half-hole based on detection-standard image edge features to obtain the burr detection result.

Images