Online Sealant Height Detection System Based on Spectral Confocal Sensor

Abstract

This invention discloses an online sealant height detection system for optical platforms based on a spectral confocal sensor, relating to the field of thickness detection technology. The system includes: defining a target area on the optical platform; acquiring height coordinate data; mapping the height coordinate data to obtain horizontal coordinate data of the target area and establishing a three-dimensional model to obtain point cloud data; mapping the point cloud data to obtain a grayscale image, inputting it into a target detection model to obtain the detection result of the target image; determining the quantity and area of ​​the detection results on the target area and calculating the flatness of the target area; if the flatness is greater than a flatness threshold, the target area is determined to be abnormal. This invention achieves sealant contour detection, improving the accuracy of sealant detection on optical platforms.

Description

Technical Field

[0001] This invention relates to the field of thickness detection technology, specifically to an online system for detecting the height of sealant on an optical platform based on a spectral confocal sensor. Background Technology

[0002] With the development of technology, thickness measurement technology, based on high-precision sensing and digital analysis, relies on ultrasonic and laser triangulation to achieve non-contact measurement. Modern systems integrate high-speed signal processing and AI algorithms, achieving micron-level accuracy. Application scenarios include online monitoring of ultra-thin materials such as lithium battery electrodes, semiconductor wafers, and metal foils, meeting the stringent requirements for thickness consistency in fields such as new energy and electronics manufacturing.

[0003] The existing technology (publication number: CN118154562A) provides an online metal surface defect detection system based on YOLO7 deep learning, which relates to the fields of industrial process inspection and metal casting production. It includes a data acquisition module, a data storage module, a data processing module, a detection module, and a display module. Compared with traditional manual quality inspection, this YOLO7-based online metal surface defect detection system proposes a brand-new online metal surface defect detection system with high real-time performance, accuracy, and digitalization. It can perform product defect detection and quality inspection by using an RGB-D camera to acquire images, completely eliminating the interference of human subjective factors, which can greatly improve the quality inspection efficiency of enterprises, improve the overall product quality, and bring strong technical strength to enterprises with low operating costs.

[0004] However, in practical applications, the RGB images obtained by photographing the sealant of the optical platform with a camera are greatly affected by light and shadow, resulting in low image extraction accuracy. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that RGB images of optical platform sealant obtained by taking pictures with a camera are greatly affected by light and shadow, resulting in low image extraction accuracy. Therefore, this invention proposes an online detection system for the height of optical platform sealant based on a spectral confocal sensor.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] First, an online detection system for sealant height on an optical platform based on a spectral confocal sensor is proposed. The system includes:

[0008] Platform data acquisition module: The optical platform defines a target area, which is any plane of the optical platform; the target area is scanned using a spectral confocal sensor to obtain height coordinate data; and the horizontal coordinate data of the target area is mapped from the height coordinate data.

[0009] 3D module: A 3D model is built based on elevation coordinate data and planar coordinate data to obtain point cloud data;

[0010] Image acquisition module: Maps point cloud data to obtain grayscale images;

[0011] Image recognition module: Identifies grayscale images as target images, uses the target images as input to the target detection model, and obtains the detection results of the target images;

[0012] Detection module: Based on the detection results of each target image, determine the number and area of ​​detection results in the target region; calculate the flatness of the target region based on the number and area of ​​detection results; if the flatness is greater than the flatness threshold, the target region is determined to be abnormal.

[0013] Optionally, the 3D module includes: a module initialization module and a point cloud module.

[0014] The module initialization module is used to acquire the 3D module and initialize the 3D module;

[0015] The point cloud module is used to analyze elevation and planar coordinate data to obtain a set of all three-dimensional coordinates of the target area; it maps the three-dimensional coordinate set to a three-dimensional module to obtain a three-dimensional model, and obtains the corresponding point cloud data based on the three-dimensional model.

[0016] Optionally, the image acquisition module may also include:

[0017] The point cloud is divided into corresponding pixel grids based on the X-axis and Y-axis data of the point cloud data;

[0018] Calculate the average of the Z-axis coordinates of the points that are divided into each pixel grid, and retain the points after averaging.

[0019] Map the points in each pixel grid to obtain the pixel values ​​of the grayscale image by mapping the Z-axis data of the points;

[0020] The corresponding grayscale image is obtained from the pixel values ​​of the grayscale image.

[0021] Optionally, the process of the object detection model includes:

[0022] The target image is used as the first feature tensor. The first feature tensor is then processed sequentially through the CBS module, CBS module, and SC_C2f module to obtain the second feature tensor. The second feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the third feature tensor. The third feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the fourth feature tensor. The fourth feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the fifth feature tensor. The fifth feature tensor is then processed sequentially through the SPPF module and GSConv module to obtain the sixth feature tensor. The third feature tensor is processed through the GSConv module to obtain the feature tensor. The third and sixth feature tensors are then imported into the Zoom_Concat module to obtain the seventh feature tensor. The third, fourth, and fifth feature tensors are then imported into the SSF module to obtain the eighth feature tensor. The seventh feature tensor is then processed sequentially through the VoVGSCSP module and GSConv module to obtain the ninth feature tensor. The second feature tensor is processed by the GSConv module to obtain a feature tensor. The third and ninth feature tensors are then imported into the Zoom_Concat module to obtain the tenth feature tensor. The tenth feature tensor is processed by the VoVGSCSP module and the GSConv module to obtain a feature tensor. The ninth feature tensor is then imported into the Concat module to obtain the eleventh feature tensor. The eleventh feature tensor is imported into the VoVGSCSP module to obtain the twelfth feature tensor. The sixth and twelfth feature tensors are processed by the GSConv module to obtain a feature tensor. The Concat module is then imported into the thirteenth feature tensor. The thirteenth feature tensor is processed by the VoVGSCSP module to obtain a feature tensor. The seventh and tenth feature tensors are processed by the VoVGSCSP module to obtain a feature tensor. The Add module is then imported into the fourteenth feature tensor. The fourteenth feature tensor is imported into the CBAM module to obtain the final feature tensor. The final feature tensor is then transmitted to the header structure for detection.

[0023] Optionally, the header structure includes: importing the final feature tensor into the Detect module to obtain the detection results of the target image.

[0024] Optionally, the structure of the SC_C2f module includes: a Split module, an SC module, and a Concat module connected in sequence;

[0025] The execution process of the SC module is as follows: the features output by the Split module are determined as input features, the input features are imported into the DW_Conv layer to obtain the first feature, the first feature is imported into two FC layers to obtain the second and third features respectively, the second and third features are fused to obtain the third feature, and the third feature is successively substituted into the FC layer, DW_Conv layer and CGLU layer to obtain the output feature.

[0026] Optionally, the operation process of the SC module includes:

[0027] ,

[0028] Where Y represents the output feature tensor of the convolution, W represents the convolution kernel, X represents the input feature tensor, and B represents the output feature tensor of the convolution. sc This represents the bias term of the convolution.

[0029] Optionally, the principles of the CGLU layer include:

[0030] The input features are segmented into a first segmentation feature and a second segmentation feature by controlling the activation mechanism. The first and second segmentation features are mapped onto weights by the Sigmoid function, and the dynamic features are obtained by element-wise multiplication with the first segmentation feature.

[0031] Optionally, the CBAM module consists of a channel attention module and a spatial attention module;

[0032] The fourteenth feature tensor obtained from the output of the Add module is determined as the original feature;

[0033] The channel attention module is used to import the original features into the max pooling layer and the average pooling layer respectively to obtain the first max pooling feature and the first average pooling feature, import the max pooling feature and the average pooling feature into the fully connected layer to obtain the fully connected feature, and import the fully connected feature into the sigmoid layer to obtain the activation feature.

[0034] The spatial attention module is used to sequentially import activation features into a max pooling layer and an average pooling layer to obtain a second max pooling feature and a second average pooling feature, respectively. The second max pooling feature and the second average pooling feature are concatenated and sequentially passed through a Conv layer and a sigmoid layer to obtain the target feature.

[0035] Optionally, the Detect module includes three detection branches. Each detection branch obtains a corresponding location and a class result through a maximum suppression algorithm. The calculation formula for the maximum suppression algorithm is as follows:

[0036] ,

[0037] ,

[0038] ,

[0039] ,

[0040] ,

[0041] Where JIoU represents the complete intersection-union ratio, JIoUi IoU represents the complete intersection-union ratio (IoU) corresponding to the i-th detector head. i This represents the crossover ratio (CROR) corresponding to the i-th detector head. Represents the predicted bounding box a and the ground truth bounding box a. gt The square of the Euclidean distance between the center points, c represents the diagonal length of the smallest bounding rectangle of the predicted box and the ground truth box, α represents the balance coefficient, and v i A represents the aspect ratio between the predicted bounding box and the ground truth bounding box corresponding to the i-th detector head; i and B i These represent the predicted bounding box and the ground truth bounding box corresponding to the i-th detector head, respectively; i represents the sequence number of the detector head. This represents the width of the ground truth bounding box corresponding to the i-th detector head. This represents the height of the ground truth bounding box corresponding to the i-th detector. This represents the width of the predicted bounding box corresponding to the i-th detector head. This represents the height of the predicted bounding box corresponding to the i-th detector head.

[0042] The beneficial effects of this invention are:

[0043] This invention proposes an online detection system for sealant height on an optical platform based on a spectral confocal sensor. The system utilizes a platform data acquisition module to: define a target area on the optical platform (any plane of the platform); scan the target area using a spectral confocal sensor to obtain height coordinate data; map the height coordinate data to obtain the horizontal coordinate data of the target area; build a 3D model based on the height and plane coordinate data to obtain point cloud data; map the point cloud data to obtain a grayscale image; define the grayscale image as the target image and use it as input to a target detection model to obtain the detection results; determine the quantity and area of ​​each detection result on the target area; calculate the flatness of the target area based on the quantity and area of ​​the detection results; if the flatness exceeds a flatness threshold, the target area is considered abnormal. By acquiring 3D data of the target area on the optical platform using a spectral confocal sensor, combined with point cloud modeling and grayscale image mapping, automated detection of sealant contour protrusions or depressions is achieved. The target detection model quickly locates the defect location and area, thereby improving detection accuracy. Attached Figure Description

[0044] Figure 1 A flowchart of an online detection system for sealant height on an optical platform based on a spectral confocal sensor, provided in an embodiment of the present invention;

[0045] Figure 2 This is a framework diagram of the target detection model provided in an embodiment of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] This invention provides an online system for detecting the height of sealant on an optical platform based on a spectral confocal sensor. See also... Figure 1 , Figure 1 A flowchart of an online sealant height detection system for an optical platform based on a spectral confocal sensor, provided in an embodiment of the present invention. The system includes the following modules:

[0048] Platform data acquisition module: Defines the target area of ​​the optical platform, which is any plane of the optical platform; scans the target area using a spectral confocal sensor to obtain height coordinate data; and maps the height coordinate data to obtain the horizontal coordinate data of the target area.

[0049] 3D module: A 3D model is built based on elevation coordinate data and planar coordinate data to obtain point cloud data;

[0050] Image acquisition module: Maps point cloud data to obtain grayscale images;

[0051] Image recognition module: Identifies grayscale images as target images, uses the target images as input to the target detection model, and obtains the detection results of the target images;

[0052] Detection module: Based on the detection results of each target image, determine the position and number of detection results in the target area; calculate the flatness of the target area based on the position and number of detection results; if the flatness is greater than the flatness threshold, the target area is determined to be abnormal.

[0053] Specifically, the optical platform is a cube, which includes multiple planes; the target image is an image of any convex or concave outline of the adhesive; the detection result is the position and area of ​​the convex or concave outline of the sealant on the target area.

[0054] Based on the online detection system for sealant height of optical platform based on spectral confocal sensor provided in this embodiment of the invention, the system acquires three-dimensional data of the target area of ​​the optical platform through spectral confocal sensor, and combines point cloud modeling and grayscale image mapping to realize automated detection of sealant contour protrusions or depressions. No contact measurement is required. The system uses target detection model to quickly locate the defect location and area, thereby improving the accuracy of detection.

[0055] In one implementation, the 3D module includes: a module initialization module and a point cloud module.

[0056] The module initialization module is used to acquire and initialize the 3D module.

[0057] Point cloud module: used to analyze elevation and planar coordinate data to obtain a set of all three-dimensional coordinates of the target area; the three-dimensional coordinate set is mapped to a three-dimensional module to obtain a three-dimensional model, and the corresponding point cloud data is obtained from the three-dimensional model.

[0058] In one implementation, a 3D model is established by fusing elevation coordinate data and planar coordinate data, outputting point cloud data, thus avoiding the depth information loss problem of traditional 2D images. The point cloud data retains the complete geometric features of the target area (such as the subtle undulations of the glue outline), providing millimeter-level spatial information for subsequent detection and significantly improving the reliability of anomaly identification.

[0059] In one implementation, the image acquisition module includes:

[0060] The point cloud is divided into corresponding pixel grids based on the X-axis and Y-axis data of the point cloud data;

[0061] Calculate the average of the Z-axis coordinates of the points that are divided into each pixel grid, and retain the points after averaging.

[0062] Map the points in each pixel grid to obtain the pixel values ​​of the grayscale image by mapping the Z-axis data of the points;

[0063] The corresponding grayscale image is obtained from the pixel values ​​of the grayscale image.

[0064] In one implementation, when mapping a 3D point cloud to a grayscale image, a conversion rule from spatial coordinates to grayscale values ​​(such as mapping height difference to grayscale gradient) is used to make the protrusions or depressions of the glue contour present high-contrast features in the image. This mapping method amplifies minute deformations, making it easier for the target detection model to capture local anomalies, thereby reducing the false negative rate.

[0065] In one implementation, the formula for obtaining pixel values ​​is:

[0066] ,

[0067] Among them, P i Z represents the pixel value. i Z represents the height value of a cloud point. s Z represents the lower limit value of the height boundary. t Z represents the upper limit of the height boundary; a Z represents the minimum height value of the sealant point cloud. b This indicates the maximum height value of the sealant dot cloud;

[0068] In one implementation, see [link to implementation details]. Figure 2 , Figure 2 This is a framework diagram of the target detection model provided in an embodiment of the present invention. The process of the target detection model includes:

[0069] The target image is used as the first feature tensor. The first feature tensor is then processed sequentially through the CBS module, CBS module, and SC_C2f module to obtain the second feature tensor. The second feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the third feature tensor. The third feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the fourth feature tensor. The fourth feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the fifth feature tensor. The fifth feature tensor is then processed sequentially through the SPPF module and GSConv module to obtain the sixth feature tensor. The third feature tensor is processed through the GSConv module to obtain the feature tensor. The third and sixth feature tensors are then imported into the Zoom_Concat module to obtain the seventh feature tensor. The third, fourth, and fifth feature tensors are then imported into the SSF module to obtain the eighth feature tensor. The seventh feature tensor is then processed sequentially through the VoVGSCSP module and GSConv module to obtain the ninth feature tensor. The second feature tensor is processed by the GSConv module to obtain a feature tensor. The third and ninth feature tensors are then imported into the Zoom_Concat module to obtain the tenth feature tensor. The tenth feature tensor is processed by the VoVGSCSP module and the GSConv module to obtain a feature tensor. The ninth feature tensor is then imported into the Concat module to obtain the eleventh feature tensor. The eleventh feature tensor is imported into the VoVGSCSP module to obtain the twelfth feature tensor. The sixth and twelfth feature tensors are processed by the GSConv module to obtain a feature tensor. The Concat module is then imported into the thirteenth feature tensor. The thirteenth feature tensor is processed by the VoVGSCSP module to obtain a feature tensor. The seventh and tenth feature tensors are processed by the VoVGSCSP module to obtain a feature tensor. The Add module is then imported into the fourteenth feature tensor. The fourteenth feature tensor is imported into the CBAM module to obtain the final feature tensor. The final feature tensor is then transmitted to the header structure for detection.

[0070] In one implementation, the header structure includes: importing the final feature tensor into the Detect module to obtain the detection result of the target image.

[0071] In one implementation, the target detection model directly takes a grayscale image as input and outputs the location and area of ​​the sealant contour anomaly on the optical platform, enabling the model to identify it more accurately, thereby improving the overall detection performance and significantly improving the accuracy of the sealant detection results on the optical platform.

[0072] In one implementation, the SC_C2f module consists of a Split module, an SC module, and a Concat module connected sequentially.

[0073] The execution process of the SC module is as follows: the features output by the Split module are determined as input features, the input features are imported into the DW_Conv layer to obtain the first feature, the first feature is imported into two FC layers to obtain the second feature and the third feature respectively, the second feature and the third feature are fused to obtain the third feature, and the third feature is sequentially substituted into the FC layer, the DW_Conv layer and the CGLU layer to obtain the output feature.

[0074] In one implementation, the operation process of the SC module includes:

[0075] ,

[0076] Where Y represents the output feature tensor of the convolution, W represents the convolution kernel, X represents the input feature tensor, and B represents the output feature tensor of the convolution. sc This represents the bias term of the convolution.

[0077] In one implementation, the principle of the CGLU layer includes:

[0078] The input features are segmented into a first segmentation feature and a second segmentation feature by controlling the activation mechanism. The first and second segmentation features are mapped onto weights by the Sigmoid function, and the dynamic features are obtained by element-wise multiplication with the first segmentation feature.

[0079] In one implementation, the SC_C2f module significantly improves the model's feature representation ability, nonlinear representation ability, and computational efficiency through the dynamic channel selection mechanism of the CGLU layer and the gradient stabilization effect of the residual connection; making the target detection model more lightweight while achieving high-precision and high-efficiency detection performance.

[0080] In one implementation, the CBAM module comprises a channel attention module and a spatial attention module;

[0081] The fourteenth feature tensor obtained from the output of the Add module is determined as the original feature;

[0082] The channel attention module is used to import the original features into the max pooling layer and the average pooling layer respectively to obtain the first max pooling feature and the first average pooling feature, import the max pooling feature and the average pooling feature into the fully connected layer to obtain the fully connected feature, and import the fully connected feature into the sigmoid layer to obtain the activation feature.

[0083] The spatial attention module is used to sequentially import activation features into a max pooling layer and an average pooling layer to obtain a second max pooling feature and a second average pooling feature, respectively. The second max pooling feature and the second average pooling feature are concatenated and sequentially passed through a Conv layer and a sigmoid layer to obtain the target feature.

[0084] In one implementation, the CBAM module adds channel attention and spatial attention mechanisms to the feature map, allowing the model to focus more on key feature information and suppress irrelevant feature information.

[0085] In one implementation, the Detect module includes three detection branches, each of which obtains a corresponding location and a category result through a maximum suppression algorithm;

[0086] In one implementation, the calculation formula for the maximum suppression algorithm is:

[0087] ,

[0088] ,

[0089] ,

[0090] ,

[0091] ,

[0092] Where JIoU represents the complete intersection-union ratio, JIoU i IoU represents the complete intersection-union ratio (IoU) corresponding to the i-th detector head. i This represents the crossover ratio (CROR) corresponding to the i-th detector head. Represents the predicted bounding box a and the ground truth bounding box a. gt The square of the Euclidean distance between the center points, c represents the diagonal length of the smallest bounding rectangle of the predicted box and the ground truth box, α represents the balance coefficient, and v i A represents the aspect ratio between the predicted bounding box and the ground truth bounding box corresponding to the i-th detector head; i and B i These represent the predicted bounding box and the ground truth bounding box corresponding to the i-th detector head, respectively; i represents the sequence number of the detector head. This represents the width of the ground truth bounding box corresponding to the i-th detector head. This represents the height of the ground truth bounding box corresponding to the i-th detector. This represents the width of the predicted bounding box corresponding to the i-th detector head. This represents the height of the predicted bounding box corresponding to the i-th detector head.

[0093] Specifically, the complete intersection-union ratio (CIU) is an indicator used to measure the degree of matching between predicted and ground truth boxes. The CIU calculates the ratio of the intersection to the union of predicted and ground truth boxes, reflecting the degree of overlap between them; the balance coefficient is obtained by staff based on historical experience.

[0094] In one implementation, based on the detection results of each target image, the number and area of ​​detection results in the target region are determined; the smoothness of the target region is calculated based on the number and area of ​​detection results; if the smoothness is greater than a smoothness threshold, the target region is determined to be abnormal, including:

[0095] The area of ​​all the detection results is used as a dataset, and the variance of the dataset is calculated to obtain the flatness. The flatness threshold is set based on historical experience.

[0096] It should be noted that, in this document, terms such as “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, system, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, system, article, or apparatus.

[0097] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. An online detection system for the height of sealant on an optical platform based on a spectral confocal sensor, characterized in that, The system includes: Platform data acquisition module: The optical platform defines a target area, which is any plane of the optical platform; the target area is scanned using a spectral confocal sensor to obtain height coordinate data; and the horizontal coordinate data of the target area is mapped from the height coordinate data. 3D module: A 3D model is built based on elevation coordinate data and planar coordinate data to obtain point cloud data; Image acquisition module: Maps point cloud data to obtain grayscale images; Image recognition module: Identifies grayscale images as target images, uses the target images as input to the target detection model, and obtains the detection results of the target images; Detection module: Based on the detection results of each target image, determine the number and area of ​​detection results in the target region; calculate the flatness of the target region based on the number and area of ​​detection results; if the flatness is greater than the flatness threshold, the target region is determined to be abnormal. The process of the target detection model includes: The target image is used as the first feature tensor. The first feature tensor is then processed sequentially through the CBS module, CBS module, and SC_C2f module to obtain the second feature tensor. The second feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the third feature tensor. The third feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the fourth feature tensor. The fourth feature tensor is then processed sequentially through the CBS module and SC_C2f module to obtain the fifth feature tensor. The fifth feature tensor is then processed sequentially through the SPPF module and GSConv module to obtain the sixth feature tensor. The third feature tensor is processed through the GSConv module to obtain the feature tensor. The third and sixth feature tensors are then imported into the Zoom_Concat module to obtain the seventh feature tensor. The third, fourth, and fifth feature tensors are then imported into the SSF module to obtain the eighth feature tensor. The seventh feature tensor is then processed sequentially through the VoVGSCSP module and GSConv module to obtain the ninth feature tensor. The second feature tensor is processed by the GSConv module to obtain feature tensors. The third and ninth feature tensors are then imported into the Zoom_Concat module to obtain the tenth feature tensor. The tenth feature tensor is processed by the VoVGSCSP module and the GSConv module to obtain feature tensors. The ninth feature tensor is then imported into the Concat module to obtain the eleventh feature tensor. The eleventh feature tensor is imported into the VoVGSCSP module to obtain the twelfth feature tensor. The sixth and twelfth feature tensors are processed by the GSConv module to obtain feature tensors. The Concat module is then imported into the thirteenth feature tensor. The thirteenth feature tensor is processed by the VoVGSCSP module to obtain feature tensors. The seventh and tenth feature tensors are processed by the VoVGSCSP module to obtain feature tensors. These are then imported into the Add module to obtain the fourteenth feature tensor. The fourteenth feature tensor is imported into the CBAM module to obtain the final feature tensor. The final feature tensor is then transmitted to the header structure for detection. The head structure includes: importing the final feature tensor into the Detect module to obtain the detection result of the target image; The structure of the SC_C2f module includes: a Split module, an SC module, and a Concat module connected in sequence; The execution process of the SC module is as follows: the features output by the Split module are determined as input features, the input features are imported into the DW_Conv layer to obtain the first feature, the first feature is imported into two FC layers to obtain the second feature and the third feature respectively, the second feature and the third feature are fused to obtain the third feature, and the third feature is sequentially substituted into the FC layer, the DW_Conv layer and the CGLU layer to obtain the output feature.

2. The online detection system for the height of sealant on an optical platform based on a spectral confocal sensor according to claim 1, characterized in that, The three-dimensional module includes: a module initialization module and a point cloud module. The module initialization module is used to acquire the 3D module and initialize the 3D module; The point cloud module is used to analyze elevation and planar coordinate data to obtain a set of all three-dimensional coordinates of the target area; it maps the three-dimensional coordinate set to a three-dimensional module to obtain a three-dimensional model, and obtains the corresponding point cloud data based on the three-dimensional model.

3. The online detection system for the height of sealant on an optical platform based on a spectral confocal sensor according to claim 1, characterized in that, The image acquisition module further includes: The point cloud is divided into corresponding pixel grids based on the X-axis and Y-axis data of the point cloud data; Calculate the average of the Z-axis coordinates of the points that are divided into each pixel grid, and retain the points after averaging. Map the points in each pixel grid to obtain the pixel values ​​of the grayscale image by mapping the Z-axis data of the points; The corresponding grayscale image is obtained from the pixel values ​​of the grayscale image.

4. The online detection system for the height of sealant on an optical platform based on a spectral confocal sensor according to claim 1, characterized in that, The operation process of the SC module includes: Y = W × X + BSC Where Y represents the output feature tensor of the convolution, W represents the convolution kernel, X represents the input feature tensor, and B represents the output feature tensor of the convolution. sc This represents the bias term of the convolution.

5. The online detection system for the height of sealant on an optical platform based on a spectral confocal sensor according to claim 1, characterized in that, The principle of the CGLU layer includes: The input features are segmented into a first segmentation feature and a second segmentation feature by controlling the activation mechanism. The first and second segmentation features are mapped onto weights by the Sigmoid function, and the dynamic features are obtained by element-wise multiplication with the first segmentation feature.

6. The online detection system for the height of sealant on an optical platform based on a spectral confocal sensor according to claim 1, characterized in that, The CBAM module comprises a channel attention module and a spatial attention module; The fourteenth feature tensor obtained from the output of the Add module is determined as the original feature; The channel attention module is used to import the original features into the max pooling layer and the average pooling layer respectively to obtain the first max pooling feature and the first average pooling feature, import the max pooling feature and the average pooling feature into the fully connected layer to obtain the fully connected feature, and import the fully connected feature into the sigmoid layer to obtain the activation feature. The spatial attention module is used to sequentially import activation features into a max pooling layer and an average pooling layer to obtain a second max pooling feature and a second average pooling feature, respectively. The second max pooling feature and the second average pooling feature are concatenated and sequentially passed through a Conv layer and a sigmoid layer to obtain the target feature.

7. The online detection system for the height of sealant on an optical platform based on a spectral confocal sensor according to claim 1, characterized in that, The Detect module includes three detection branches, each of which obtains a corresponding location and a category result through a maximum value suppression algorithm.

Images