Copper wire quality online detection method and system

Abstract

The invention relates to the technical field of online detection, in particular to a copper wire quality online detection method and system. A copper wire quality on-line detection system comprises a data acquisition module, a copper wire reflection intensity prediction module, a light source parameter adjustment module and a copper wire defect detection module. By constructing the copper wire surface reflection thermodynamic diagram, the reflection intensity distribution of the copper wire surface is predicted, and the light source parameters are adjusted based on the copper wire surface reflection thermodynamic diagram, so that the influence of high reflection of the copper wire surface on image recognition is weakened; the copper wire surface image is identified based on the copper wire surface reflection thermodynamic diagram and the copper wire defect detection model, the copper wire quality condition is detected, the copper wire surface reflection thermodynamic diagram can serve as priori knowledge to provide reference for subsequent copper wire image identification, and the accuracy of copper wire quality detection is improved.

Description

Technical Field

[0001] The present invention relates to the field of on-line detection technology, and particularly relates to an on-line detection method and system for the quality of copper wires. Background Art

[0002] As an important material in the fields of electronics, electric power and communication, the quality of copper wires directly affects the performance and reliability of products. Therefore, on-line detection technology plays a key role in modern manufacturing. Traditional methods for detecting the quality of copper wires mainly rely on machine vision systems to identify surface defects, such as scratches, cracks or non-uniformities, through image acquisition and processing. However, the surface of copper wires has a highly reflective characteristic, resulting in strong specular reflection and scattering of light on the surface, which has a significant impact on image recognition of defects. Specifically, the high reflectivity causes glare, shadows and reflection noise in the image, interfering with the accuracy of feature extraction algorithms. For example, edge detection and texture analysis algorithms are difficult to distinguish real defects from reflection effects, thus increasing the false alarm rate and reducing the accuracy of defect detection. At the same time, reflection interference may cause image distortion or information loss. Especially in a dynamic production line environment, changes in the light source further amplify these problems, resulting in low detection efficiency and an increased risk of misjudgment. Existing technologies, such as vision systems based on a single light source or simple filtering algorithms, although can partially alleviate the reflection problem, still cannot achieve real-time and high-precision defect recognition, ultimately restricting the overall efficiency of the production line and product quality control. Summary of the Invention

[0003] The present invention collects copper wire surface point cloud data of the copper wire entering the detection area, and collects light source light intensity data and light source angle data at the light source to construct light source feature sequence data. Then, through a copper wire reflection intensity prediction model, the copper wire surface point cloud data and the light source feature sequence data are processed to construct a copper wire surface reflection heat map, predict the reflection intensity distribution on the copper wire surface, and adjust the light source parameters based on the copper wire surface reflection heat map to weaken the influence of high reflection on the copper wire surface on image recognition. Also, based on the copper wire surface reflection heat map and a copper wire defect detection model, image recognition is performed on the copper wire surface image to detect the quality of the copper wire. The copper wire surface reflection heat map can be used as prior knowledge to provide reference for subsequent image recognition of copper wires, improving the accuracy of copper wire quality detection.

[0004] The present invention provides an on-line detection method for the quality of copper wires, including:

[0005] Obtain copper wire surface point cloud data of the copper wire surface, obtain light source light intensity data at the light source and light source angle data corresponding to the light source, and form light source feature sequence data by combining the light source feature data corresponding to the current time stamp with the light source feature data corresponding to the previous N-1 time stamps;

[0006] Send the point cloud data of the copper wire surface and the light source feature sequence data into the copper wire reflection intensity prediction model for processing, and output the copper wire surface reflection heat map;

[0007] Construct a light source parameter adjustment scheme based on the copper wire surface reflection heat map, and adjust the light source based on the light source parameter adjustment scheme;

[0008] After adjusting the light source through the light source parameter adjustment scheme, collect an image of the copper wire surface to obtain the copper wire surface image, and then send the copper wire surface image and the copper wire surface reflection heat map into the copper wire defect detection model for processing to obtain copper wire defect information;

[0009] The copper wire reflection intensity prediction model includes a spatial feature extraction layer, a light source time series feature extraction layer, a feature splicing layer, and a reflection prediction layer. Among them, the spatial feature extraction layer is used to process the point cloud data of the copper wire surface and construct the copper wire surface feature data; the light source time series feature extraction layer is used to perform time series analysis on the light source feature sequence data and construct the light source time series feature data; the feature splicing layer is used to splice the copper wire surface feature data and the light source time series feature data at the head and tail to construct the copper wire reflection intensity analysis vector; the reflection prediction layer is used to perform a deconvolution operation on the copper wire reflection intensity analysis vector to construct the copper wire surface reflection heat map.

[0010] As a preferred aspect, sending the copper wire surface image and the copper wire surface reflection heat map into the copper wire defect detection model for processing to obtain copper wire defect information specifically includes the following content:

[0011] The copper wire defect detection model includes a defect enhancement layer and a defect detection layer. Among them, the defect enhancement layer is used to perform a self-attention mechanism operation on the copper wire surface image based on the copper wire surface reflection heat map to construct the copper wire surface enhanced image; the defect detection layer is established based on the YOLOV5 model and is used to process the copper wire surface enhanced image and output the copper wire defect information;

[0012] Performing a self-attention mechanism operation on the copper wire surface image based on the copper wire surface reflection heat map through the defect enhancement layer to construct the copper wire surface enhanced image specifically includes the following content: performing matrix multiplication operations on the copper wire surface image with the value weight matrix and the key weight matrix respectively to construct the corresponding copper wire image value vector V and the copper wire surface key vector K, performing matrix multiplication operation on the copper wire surface reflection heat map with the query weight matrix to construct the copper wire surface query vector Q, and implementing the self-attention mechanism operation through the following formula: G = softmax(QK T / D 0.5 )V, where G is the copper wire surface enhanced image, T is the matrix transpose operation, and D is the dimension size of the copper wire surface key vector K.

[0013] As a preferred aspect, a light source parameter adjustment scheme is constructed based on the copper wire surface reflection thermal map, which specifically includes the following steps:

[0014] The copper wire surface reflection thermal map is sent into the light source parameter adjustment network for processing, and a light source parameter adjustment scheme is output. The light source parameter adjustment network includes a feature extraction layer and a light source parameter adjustment scheme output layer. The feature extraction layer is used to extract features from the copper wire surface reflection thermal map to construct the copper wire surface reflection distribution features; the light source parameter adjustment network is used to analyze the copper wire surface reflection distribution features to construct the corresponding light source parameter adjustment scheme.

[0015] As a preferred aspect, the copper wire reflection intensity prediction model is trained, which specifically includes the following steps:

[0016] Obtain the copper wire reflection intensity prediction training samples. The copper wire reflection intensity prediction training samples include the copper wire surface point cloud data and the light source feature sequence data. The copper wire reflection intensity prediction training samples are labeled through the copper wire surface reflection thermal map. All the labeled copper wire reflection intensity prediction training samples are composed into the copper wire reflection intensity prediction training set. The copper wire reflection intensity prediction model is trained through the copper wire reflection intensity prediction training set, and the training target is the labeled copper wire surface reflection thermal map.

[0017] As a preferred aspect, the copper wire defect detection model is trained, which specifically includes the following steps:

[0018] Obtain the copper wire defect detection training samples. The copper wire defect detection training samples include the copper wire surface image and the copper wire surface reflection thermal map. The copper wire defect detection training samples are labeled through the copper wire defect information. All the labeled copper wire defect detection training samples are composed into the copper wire defect detection training set. The copper wire defect detection model is trained through the copper wire defect detection training set, and the training target is the labeled copper wire defect information.

[0019] As a preferred aspect, the light source parameter adjustment network is trained, which specifically includes the following steps:

[0020] Obtain the light source parameter adjustment training samples. The light source parameter adjustment training samples include the copper wire surface reflection thermal map. The light source parameter adjustment training samples are labeled through the light source parameter adjustment scheme. All the labeled light source parameter adjustment training samples are composed into the light source parameter adjustment training set. The light source parameter adjustment network is trained through the light source parameter adjustment training set, and the training target is the labeled light source parameter adjustment scheme.

[0021] The present invention also provides an on-line copper wire quality detection system, including:

[0022] The data acquisition module is used to obtain the copper wire surface point cloud data on the copper wire surface, obtain the light source intensity data at the light source and the light source angle data corresponding to the light source, and form the light source feature sequence data by combining the light source feature data corresponding to the current time stamp with the light source feature data corresponding to the previous N - 1 time stamps;

[0023] The copper wire reflection intensity prediction module is used to send the copper wire surface point cloud data and the light source feature sequence data into the copper wire reflection intensity prediction model for processing, and output the copper wire surface reflection heat map;

[0024] The light source parameter adjustment module is used to construct a light source parameter adjustment scheme based on the copper wire surface reflection heat map, and adjust the light source based on the light source parameter adjustment scheme;

[0025] The copper wire defect detection module is used to, after adjusting the light source through the light source parameter adjustment scheme, collect an image of the copper wire surface to obtain the copper wire surface image, and then send the copper wire surface image and the copper wire surface reflection heat map into the copper wire defect detection model for processing to obtain the copper wire defect information;

[0026] The copper wire reflection intensity prediction model includes a spatial feature extraction layer, a light source time series feature extraction layer, a feature splicing layer, and a reflection prediction layer. Among them, the spatial feature extraction layer is used to process the copper wire surface point cloud data to construct copper wire surface feature data; the light source time series feature extraction layer is used to perform time series analysis on the light source feature sequence data to construct light source time series feature data; the feature splicing layer is used to splice the copper wire surface feature data and the light source time series feature data head - to - tail to construct a copper wire reflection intensity analysis vector; the reflection prediction layer is used to perform a deconvolution operation on the copper wire reflection intensity analysis vector to construct the copper wire surface reflection heat map.

[0027] The present invention has the following advantages:

[0028] The present invention collects the copper wire surface point cloud data of the copper wire entering the detection area, and collects the light source intensity data and the light source angle data at the light source, constructs the light source feature sequence data, then processes the copper wire surface point cloud data and the light source feature sequence data through the copper wire reflection intensity prediction model to construct the copper wire surface reflection heat map, predicts the reflection intensity distribution on the copper wire surface, adjusts the light source parameters based on the copper wire surface reflection heat map to weaken the influence of high reflection on the copper wire surface on image recognition, and also performs image recognition on the copper wire surface image based on the copper wire surface reflection heat map and the copper wire defect detection model to detect the quality of the copper wire. The copper wire surface reflection heat map can be used as prior knowledge to provide reference for subsequent image recognition of the copper wire, improving the accuracy of copper wire quality detection. Brief Description of the Drawings

[0029] Figure 1This is a schematic structural diagram of the on-line copper wire quality detection system adopted in the embodiments of the present invention. Specific embodiments

[0030] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention.

[0031] Embodiment 1, an on-line copper wire quality detection method, including:

[0032] During the process of performing copper wire quality detection, whenever a copper wire enters the detection area, the copper wire surface point cloud data on the surface of the copper wire is obtained through a ToF sensor arranged in the detection area. The copper wire surface point cloud data here includes the X, Y, and Z coordinates and light intensity information of several points, which can reflect the characteristic information on the surface of the copper wire. At the same time, the light source light intensity data at the light source is obtained through a light sensor arranged at the light source, and the light source angle data corresponding to the light source is obtained through an inertial measurement unit arranged at the light source. Here, it should be noted that the light source angle refers to the angle between the light source orientation and the ground, including the pitch angle, yaw angle, or roll angle. The light source light intensity data and the light source angle data together reflect the characteristic information of the light rays emitted by the light source. The light source characteristic data is composed of the light source light intensity data and the light source angle data corresponding to the current timestamp. Moreover, the light sensor and the inertial measurement unit perform data acquisition tasks according to the set acquisition frequency. Therefore, the light source characteristic data corresponding to the current timestamp and the light source characteristic data corresponding to the previous N-1 timestamps are composed of light source characteristic sequence data. The light source characteristic sequence data reflects the intensity time series change (due to the influence of the internal voltage of the light source) and the angle time series change (due to the mechanical vibration of the production line) of the light rays emitted by the light source. The intensity time series change and the angle time series change of the light rays have an impact on the reflection situation on the surface of the copper wire;

[0033] The point cloud data of the copper wire surface and the light source feature sequence data are sent into the copper wire reflection intensity prediction model for processing, and a copper wire surface reflection heat map is output. The copper wire surface reflection heat map is a 2D image used to represent the predicted reflection intensity distribution on the copper wire surface. Since the copper wire is a highly reflective material and mainly exhibits specular reflection, if the light source is stable, the images collected for the copper wire may be clear. However, once the light source fluctuates (due to changes in the internal voltage of the light source or mechanical vibrations on the production line), the high reflectivity of the copper wire material will amplify the change in reflectivity caused by this light source fluctuation. In this case, when collecting images of the copper wire, stripes may appear in the images corresponding to the normal areas of the copper wire, while bright spots will be formed due to overexposure in the areas corresponding to the defective areas of the copper wire. This will have a great impact on the subsequent quality inspection of the copper wire through image recognition technology. By predicting the reflection intensity distribution on the copper wire surface, the reflectivity on the copper wire surface can be analyzed in advance, and then the light source parameters can be adjusted to weaken the influence of the high reflection on the copper wire surface on image recognition. And since the areas with high reflection may be defective areas of the copper wire or natural bends of the copper wire, the copper wire surface reflection heat map can also provide reference for subsequent image recognition of the copper wire as prior knowledge;

[0034] Based on the copper wire surface reflection heat map, a light source parameter adjustment scheme is constructed, and the light source is adjusted based on the light source parameter adjustment scheme. Irradiating the copper wire with the adjusted light source can weaken the high reflection on the copper wire surface and improve the quality of subsequent image acquisition;

[0035] After adjusting the light source according to the light source parameter adjustment scheme, the surface of the copper wire is imaged by an industrial camera to obtain a copper wire surface image. Then, the copper wire surface image and the copper wire surface reflection heat map are sent into the copper wire defect detection model for processing to obtain copper wire defect information. Here, the copper wire defect information includes two categories. One is that no copper wire defect is found, that is, the copper wire is of qualified quality, and the copper wire defect information is expressed as a defect-free label. The other is that a copper wire defect is found and the copper wire quality is unqualified. At this time, the copper wire defect information is expressed as the copper wire defect type and the copper wire defect annotation box (the location where the copper wire defect is located) to facilitate the operator to find the cause of the copper wire defect;

[0036] The copper wire reflection intensity prediction model includes a spatial feature extraction layer, a light source temporal feature extraction layer, a feature splicing layer, and a reflection prediction layer. The spatial feature extraction layer is used to process the point cloud data on the surface of the copper wire to construct the surface feature data of the copper wire. It should be noted that the spatial feature extraction layer is established based on the PointNet model, specifically the PointNet model with the last layer (classification layer or segmentation layer) removed, which is used to extract the spatial features on the surface of the copper wire. The extracted surface feature data of the copper wire can reflect the path and intensity distribution of light reflection, improving the accuracy of predicting the reflection intensity distribution on the surface of the copper wire. The light source temporal feature extraction layer is used to perform temporal analysis on the light source feature sequence data to construct the light source temporal feature data, which reflects the fluctuation changes of the light source and affects the reflection intensity distribution on the surface of the copper wire. The feature splicing layer is used to splice the surface feature data of the copper wire and the light source temporal feature data head-to-tail to construct the copper wire reflection intensity analysis vector. The reflection prediction layer is used to perform a deconvolution operation on the copper wire reflection intensity analysis vector to construct the reflection heat map on the surface of the copper wire. Here, the reflection prediction layer is established based on the decoder in the U-net model and includes several deconvolution operations.

[0037] In this application, the point cloud data on the surface of the copper wire entering the detection area is collected, and the light intensity data and light source angle data at the light source are collected to construct the light source feature sequence data. Then, the copper wire reflection intensity prediction model is used to process the point cloud data on the surface of the copper wire and the light source feature sequence data to construct the reflection heat map on the surface of the copper wire, predict the reflection intensity distribution on the surface of the copper wire, and adjust the light source parameters based on the reflection heat map on the surface of the copper wire to weaken the influence of high reflection on the surface of the copper wire on image recognition. Also, based on the reflection heat map on the surface of the copper wire and the copper wire defect detection model, image recognition is performed on the surface image of the copper wire to detect the quality of the copper wire. The reflection heat map on the surface of the copper wire can provide a reference for subsequent image recognition of the copper wire as prior knowledge, improving the accuracy of copper wire quality detection.

[0038] The surface image of the copper wire and the reflection heat map on the surface of the copper wire are sent into the copper wire defect detection model for processing to obtain the copper wire defect information, which specifically includes the following content:

[0039] The copper wire defect detection model includes a defect enhancement layer and a defect detection layer. The defect enhancement layer is used to perform a self-attention mechanism operation on the copper wire surface image based on the copper wire surface reflection heat map to construct an enhanced image of the copper wire surface. Since the area with high reflection may be the defect area of the copper wire, enhancing the features corresponding to the high-reflection area of the copper wire surface image can make the detection of the copper wire surface image pay more attention to the areas where defects may occur and improve the accuracy of copper wire defect detection. The defect detection layer is established based on the YOLOV5 model and is used to process the enhanced image of the copper wire surface and output copper wire defect information. The YOLOV5 model is a commonly used neural network model for image recognition and can output the bounding box, confidence level, and class probability of the defect based on the image.

[0040] The defect enhancement layer performs a self-attention mechanism operation on the copper wire surface image based on the copper wire surface reflection heat map to construct an enhanced image of the copper wire surface, which specifically includes the following content: performing matrix multiplication operations on the copper wire surface image with the value weight matrix and the key weight matrix respectively to construct the corresponding copper wire image value vector V and the copper wire surface key vector K, performing matrix multiplication operation on the copper wire surface reflection heat map with the query weight matrix to construct the copper wire surface query vector Q, and implementing the self-attention mechanism operation through the following formula: G = softmax(QK T / D 0.5 )V, where G is the enhanced image of the copper wire surface, T is the matrix transpose operation, and D is the dimension size of the copper wire surface key vector K. It should be noted that the self-attention mechanism operation here is implemented with reference to the Transformer model, and the value weight matrix, the key weight matrix, and the query weight matrix are all adjustable parameters, and the parameters are adjusted following the end-to-end training of the copper wire defect detection model.

[0041] Construct a light source parameter adjustment scheme based on the copper wire surface reflection heat map, which specifically includes the following steps:

[0042] Send the copper wire surface reflection heat map into the light source parameter adjustment network for processing to output the light source parameter adjustment scheme. The light source parameter adjustment network includes a feature extraction layer and a light source parameter adjustment scheme output layer. The feature extraction layer is constructed based on a convolutional neural network. In this embodiment, the VGG16 model is used, specifically, the last fully connected layer of the VGG16 model is removed, which is used to extract features from the copper wire surface reflection heat map and construct the copper wire surface reflection distribution features. The light source parameter adjustment network is established based on a BP neural network and is used to analyze the copper wire surface reflection distribution features and construct the corresponding light source parameter adjustment scheme.

[0043] Train the copper wire reflection intensity prediction model, which specifically includes the following steps:

[0044] Obtain the training samples for predicting the reflection intensity of copper wire. The training samples for predicting the reflection intensity of copper wire include the point cloud data on the surface of the copper wire and the light source feature sequence data. It should be noted that the point cloud data on the surface of the copper wire and the light source feature sequence data in the training samples for predicting the reflection intensity of copper wire are obtained from the actual production line, or can be obtained by constructing a simulation environment through simulation software (such as Blender and Unity, etc.). Label the training samples for predicting the reflection intensity of copper wire with the thermal reflection map on the surface of the copper wire. Here, the thermal reflection map on the surface of the copper wire for labeling is also obtained by the operator according to the actual production line or calculated through simulation software. Combine all the labeled training samples for predicting the reflection intensity of copper wire into a training set for predicting the reflection intensity of copper wire. Train the prediction model for the reflection intensity of copper wire with the training set for predicting the reflection intensity of copper wire. The training target is the labeled thermal reflection map on the surface of the copper wire. Determine whether the training conditions are met. The training conditions can be that the accuracy rate of the prediction model for the reflection intensity of copper wire meets the expectation. If the accuracy rate of the prediction model for the reflection intensity of copper wire meets the expectation, output the trained prediction model for the reflection intensity of copper wire; otherwise, continue to train the prediction model for the reflection intensity of copper wire with the training set for predicting the reflection intensity of copper wire.

[0045] Train the copper wire defect detection model, which specifically includes the following steps:

[0046] Obtain the training samples for copper wire defect detection. The training samples for copper wire defect detection include the surface image of the copper wire and the thermal reflection map on the surface of the copper wire. It should be noted that the surface image of the copper wire and the thermal reflection map on the surface of the copper wire here are obtained at the same time stamp, and are also obtained according to the actual production line or through simulation software. Label the training samples for copper wire defect detection with the copper wire defect information. Here, the labeled copper wire defect information is obtained by the operator using a labeling tool (such as LabelStudio or RectLabel) to label the defects. Combine all the labeled training samples for copper wire defect detection into a training set for copper wire defect detection. Train the copper wire defect detection model with the training set for copper wire defect detection. The training target is the labeled copper wire defect information. Determine whether the training conditions are met. The training conditions can be that the accuracy rate of the copper wire defect detection model meets the expectation. If the accuracy rate of the copper wire defect detection model meets the expectation, output the trained copper wire defect detection model; otherwise, continue to train the copper wire defect detection model with the training set for predicting the reflection intensity of copper wire.

[0047] Train the light source parameter adjustment network, which specifically includes the following steps:

[0048] Obtain a training sample for light source parameter adjustment. The training sample for light source parameter adjustment includes a thermal map of the reflection on the surface of the copper wire. Here, the thermal map of the reflection on the surface of the copper wire has the same source as the thermal map of the reflection on the surface of the copper wire during the training of the copper wire reflection intensity prediction model. Label the training sample for light source parameter adjustment through a light source parameter adjustment scheme. Here, the labeling process is the optimal light source adjustment scheme obtained by the operator according to the feedback effect after the light source adjustment. Combine all the labeled training samples for light source parameter adjustment to form a training set for light source parameter adjustment. Train the light source parameter adjustment network through the training set for light source parameter adjustment. The training objective is the labeled light source parameter adjustment scheme. Determine whether the training conditions are met. The training condition is that the accuracy of the light source parameter adjustment network meets the expectation. If the accuracy of the light source parameter adjustment network meets the expectation, output the trained light source parameter adjustment network; otherwise, continue to train the light source parameter adjustment network through the training set for copper wire reflection intensity prediction.

[0049] Embodiment 2. An on-line copper wire quality detection system. Refer to Figure 1 , including:

[0050] A data acquisition module, which is used to obtain the copper wire surface point cloud data on the surface of the copper wire through the ToF sensor set in the detection area whenever the copper wire enters the detection area during the process of performing copper wire quality detection. Here, the copper wire surface point cloud data includes the X, Y, and Z coordinates and light intensity information of several points, which can reflect the characteristic information of the copper wire surface, and; at the same time, obtain the light source light intensity data at the light source through the light sensor set at the light source, and obtain the light source angle data corresponding to the light source through the inertial measurement unit set at the light source. Here, it should be noted that the light source angle refers to the angle between the light source orientation and the ground, including the pitch angle, yaw angle, or roll angle. The light source light intensity data and the light source angle data together reflect the characteristic information of the light emitted by the light source. Combine the light source light intensity data and the light source angle data corresponding to the current timestamp to form light source characteristic data. Moreover, the light sensor and the inertial measurement unit perform the data acquisition task according to the set acquisition frequency. Therefore, combine the light source characteristic data corresponding to the current timestamp with the light source characteristic data corresponding to the previous N - 1 timestamps to form light source characteristic sequence data. The light source characteristic sequence data reflects the intensity temporal variation (due to the influence of the internal voltage of the light source) and the angular temporal variation (due to the mechanical vibration of the production line) of the light emitted by the light source. The intensity temporal variation and the angular temporal variation of the light have an impact on the reflection situation on the surface of the copper wire;

[0051] The copper wire reflection intensity prediction module is used to send the point cloud data of the copper wire surface and the light source feature sequence data into the copper wire reflection intensity prediction model for processing, and output the copper wire surface reflection heat map. The copper wire surface reflection heat map is a 2D image, which is used to represent the predicted reflection intensity distribution on the copper wire surface. Since the copper wire is a highly reflective material and mainly undergoes specular reflection, if the light source is stable, the images collected for the copper wire may be clear. However, once the light source fluctuates (such as changes in the internal voltage of the light source or mechanical vibrations of the production line), the high reflectivity of the copper wire material will amplify the change in reflectivity caused by this light source fluctuation. In this case, when collecting images of the copper wire, stripes may appear in the images corresponding to the normal areas of the copper wire, while bright spots will be formed due to overexposure in the areas corresponding to the defective areas of the copper wire. This will have a great impact on the subsequent quality inspection of the copper wire through image recognition technology. By predicting the reflection intensity distribution on the copper wire surface, the reflectivity of the copper wire surface can be analyzed in advance, and then the light source parameters can be adjusted to weaken the influence of the high reflection on the copper wire surface on image recognition. And since the areas with high reflection may be defective areas of the copper wire or natural bends of the copper wire, the copper wire surface reflection heat map can also provide reference for subsequent image recognition of the copper wire as prior knowledge;

[0052] The light source parameter adjustment module is used to construct a light source parameter adjustment scheme based on the copper wire surface reflection heat map, and adjust the light source based on the light source parameter adjustment scheme. By irradiating the copper wire with the adjusted light source, the high reflection situation on the copper wire surface can be weakened, and the quality of subsequent image collection can be improved;

[0053] The copper wire defect detection module is used to, after adjusting the light source according to the light source parameter adjustment scheme, collect images of the copper wire surface through an industrial camera to obtain the copper wire surface image, and then send the copper wire surface image and the copper wire surface reflection heat map into the copper wire defect detection model for processing to obtain the copper wire defect information. Here, the copper wire defect information includes two categories. One is that no copper wire defect is found, that is, the copper wire quality is qualified, and the copper wire defect information is represented as a defect-free label. The other is that a copper wire defect is found and the copper wire quality is unqualified. At this time, the copper wire defect information is represented as the copper wire defect type and the copper wire defect annotation box (the location where the copper wire defect is located), so as to facilitate the operator to find the reason for the occurrence of the copper wire defect;

[0054] The copper wire reflection intensity prediction model includes a spatial feature extraction layer, a light source temporal feature extraction layer, a feature splicing layer, and a reflection prediction layer. The spatial feature extraction layer is used to process the point cloud data on the copper wire surface to construct the copper wire surface feature data. It should be noted that the spatial feature extraction layer is established based on the PointNet model, specifically the PointNet model with the last layer (classification layer or segmentation layer) removed, which is used to extract the spatial features of the copper wire surface. The extracted copper wire surface feature data can reflect the path and intensity distribution of light reflection, improving the accuracy of predicting the reflection intensity distribution on the copper wire surface. The light source temporal feature extraction layer is used to perform temporal analysis on the light source feature sequence data to construct the light source temporal feature data, which reflects the fluctuation changes of the light source and affects the reflection intensity distribution on the copper wire surface. The feature splicing layer is used to splice the copper wire surface feature data and the light source temporal feature data end to end to construct the copper wire reflection intensity analysis vector. The reflection prediction layer is used to perform a deconvolution operation on the copper wire reflection intensity analysis vector to construct the copper wire surface reflection heat map. Here, the reflection prediction layer is established based on the decoder in the U-net model and includes several deconvolution operations.

[0055] It should be understood that those of ordinary skill in the art can make improvements or transformations according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention. The parts not described in detail in this specification belong to the prior art well-known to those skilled in the art.

Claims

1. An on-line detection method for the quality of copper wire, characterized in that Including: Obtain the point cloud data of the copper wire surface, obtain the light intensity data of the light source at the light source and the light source angle data corresponding to the light source, and form the light source feature sequence data by combining the light source feature data corresponding to the current timestamp with the light source feature data corresponding to the previous N - 1 timestamps; Send the copper wire surface point cloud data and the light source feature sequence data into the copper wire reflection intensity prediction model for processing, and output the copper wire surface reflection heat map; Construct a light source parameter adjustment scheme based on the copper wire surface reflection heat map, and adjust the light source based on the light source parameter adjustment scheme; after adjusting the light source through the light source parameter adjustment scheme, collect an image of the copper wire surface to obtain the copper wire surface image, and then send the copper wire surface image and the copper wire surface reflection heat map into the copper wire defect detection model for processing to obtain the copper wire defect information; The copper wire reflection intensity prediction model includes a spatial feature extraction layer, a light source time series feature extraction layer, a feature splicing layer, and a reflection prediction layer. Among them, the spatial feature extraction layer is used to process the copper wire surface point cloud data to construct the copper wire surface feature data; the light source time series feature extraction layer is used to perform time series analysis on the light source feature sequence data to construct the light source time series feature data; The feature splicing layer is used to splice the copper wire surface feature data and the light source time series feature data end to end to construct the copper wire reflection intensity analysis vector; the reflection prediction layer is used to perform a deconvolution operation on the copper wire reflection intensity analysis vector to construct the copper wire surface reflection heat map.

2. The on-line detection method for the quality of copper wire according to claim 1, characterized in that Send the copper wire surface image and the copper wire surface reflection heat map into the copper wire defect detection model for processing to obtain the copper wire defect information, which specifically includes the following content: the copper wire defect detection model includes a defect enhancement layer and a defect detection layer. Among them, the defect enhancement layer is used to perform a self-attention mechanism operation on the copper wire surface image based on the copper wire surface reflection heat map to construct the copper wire surface enhanced image; the defect detection layer is established based on the YOLOV5 model and is used to process the copper wire surface enhanced image to output the copper wire defect information; Perform self-attention mechanism operations on the copper wire surface image based on the copper wire surface reflection heat map through the defect enhancement layer to construct an enhanced image of the copper wire surface, which specifically includes the following: perform matrix multiplication operations on the copper wire surface image with the value weight matrix and the key weight matrix respectively to construct the corresponding copper wire image value vector V and the copper wire surface key vector K, perform matrix multiplication operations on the copper wire surface reflection heat map with the query weight matrix to construct the copper wire surface query vector Q, and implement the self-attention mechanism operation through the following formula: G = softmax(QK T / D 0.5 )V, where G is the enhanced image of the copper wire surface, T is the matrix transpose operation, and D is the dimension size of the copper wire surface key vector K.

3. The on-line detection method for the quality of copper wires according to claim 2, characterized in that, Construct a light source parameter adjustment scheme based on the copper wire surface reflection heat map, which specifically includes the following steps: Send the copper wire surface reflection heat map into the light source parameter adjustment network for processing to output the light source parameter adjustment scheme. The light source parameter adjustment network includes a feature extraction layer and a light source parameter adjustment scheme output layer. Among them, the feature extraction layer is used to extract features from the copper wire surface reflection heat map to construct the copper wire surface reflection distribution feature; the light source parameter adjustment network is used to analyze the copper wire surface reflection distribution feature to construct the corresponding light source parameter adjustment scheme.

4. The on-line detection method for the quality of copper wire according to claim 3, characterized in that, Train the copper wire reflection intensity prediction model, which specifically includes the following steps: Obtain the copper wire reflection intensity prediction training samples, which include the copper wire surface point cloud data and the light source feature sequence data in the copper wire reflection intensity prediction training samples. Label the copper wire reflection intensity prediction training samples through the copper wire surface reflection heat map, and form the copper wire reflection intensity prediction training set by all the labeled copper wire reflection intensity prediction training samples. Train the copper wire reflection intensity prediction model through the copper wire reflection intensity prediction training set, and the training target is the labeled copper wire surface reflection heat map.

5. The on-line detection method for the quality of copper wire according to claim 4, characterized in that, Train the copper wire defect detection model, which specifically includes the following steps: Obtain copper wire defect detection training samples. The copper wire defect detection training samples include copper wire surface images and copper wire surface reflection thermal maps. Label the copper wire defect detection training samples based on copper wire defect information. Combine all the labeled copper wire defect detection training samples to form a copper wire defect detection training set. Train the copper wire defect detection model using the copper wire defect detection training set, with the training target being the labeled copper wire defect information.

6. The on-line detection method for the quality of copper wire according to claim 5, characterized in that Train the light source parameter adjustment network, which specifically includes the following steps: Obtain light source parameter adjustment training samples. The light source parameter adjustment training samples include copper wire surface reflection thermal maps. Label the light source parameter adjustment training samples according to the light source parameter adjustment scheme. Combine all the labeled light source parameter adjustment training samples to form a light source parameter adjustment training set. Train the light source parameter adjustment network using the light source parameter adjustment training set, with the training target being the labeled light source parameter adjustment scheme.

7. An on-line copper wire quality detection system, characterized in that, The system applies the on-line copper wire quality detection method described in any one of claims 1-6 above, including: A data acquisition module for obtaining copper wire surface point cloud data on the copper wire surface, obtaining light source light intensity data at the light source and light source angle data corresponding to the light source, and combining the light source feature data corresponding to the current timestamp with the light source feature data corresponding to the previous N-1 timestamps to form light source feature sequence data; A copper wire reflection intensity prediction module for sending the copper wire surface point cloud data and the light source feature sequence data into the copper wire reflection intensity prediction model for processing and outputting a copper wire surface reflection thermal map; A light source parameter adjustment module for constructing a light source parameter adjustment scheme based on the copper wire surface reflection thermal map and adjusting the light source based on the light source parameter adjustment scheme; A copper wire defect detection module for, after adjusting the light source according to the light source parameter adjustment scheme, collecting an image of the copper wire surface to obtain a copper wire surface image, and then sending the copper wire surface image and the copper wire surface reflection thermal map into the copper wire defect detection model for processing to obtain copper wire defect information; The copper wire reflection intensity prediction model includes a spatial feature extraction layer, a light source time series feature extraction layer, a feature splicing layer, and a reflection prediction layer. Among them, the spatial feature extraction layer is used to process the copper wire surface point cloud data to construct copper wire surface feature data; the light source time series feature extraction layer is used to perform time series analysis on the light source feature sequence data to construct light source time series feature data; the feature splicing layer is used to splice the copper wire surface feature data and the light source time series feature data head to tail to construct a copper wire reflection intensity analysis vector; the reflection prediction layer is used to perform a deconvolution operation on the copper wire reflection intensity analysis vector to construct a copper wire surface reflection thermal map.

Images