LED color difference detection system based on computer vision

Abstract

The invention provides an LED color difference detection system based on computer vision, which comprises an image acquisition module, a preprocessing module, a detection analysis module and a result output module, the preprocessing module adopts an attention-oriented denoising convolutional neural network ADNet to carry out denoising, enhancement and other processing on an image, the detection analysis module utilizes a multi-scale enhanced YOLOv8 target detection algorithm to carry out chromatic aberration analysis on the processed image, and the result output module displays a detection result to a user in a visual form. The LED board image is subjected to denoising and enhancement processing by using an image processing technology, and feature extraction and analysis are performed on the LED board image through a deep learning algorithm based on computer vision, so that efficient acquisition and accurate color difference analysis of the LED board image are realized, the problems of low efficiency and high false detection rate of a traditional LED color difference detection method are solved, and the detection accuracy of the LED board image is improved. The production quality of the LED board can be obviously improved, and the production cost is reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of computer vision and image processing, and in particular relates to a LED color difference detection and analysis system based on computer vision. Background Art

[0002] With the rapid development of LED lighting technology, its high efficiency, long life and environmental protection make LEDs increasingly widely used in various fields such as daily life, commercial lighting and industrial production. However, LEDs may cause color deviations due to various reasons during the production process, affecting product quality and user experience. Traditional LED color difference detection methods mostly rely on manual detection. Although this method is intuitive and simple, it has many limitations. Under large-scale production, manual detection has problems such as low efficiency and high false detection rate. Therefore, it is of great significance to develop an efficient and accurate LED color difference detection and analysis system. Summary of the invention

[0003] The purpose of the present invention is to provide a LED color difference detection and analysis system based on computer vision to solve the problems of low efficiency and high false detection rate of traditional LED color difference detection methods.

[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a computer vision-based LED color difference detection system, the system comprising: an image acquisition module, used to obtain an LED board color difference image data set; a preprocessing module, used to perform denoising, enhancement and other processing on the LED board color difference data set image to improve the detection accuracy; a detection and analysis module, using a multi-scale feature enhanced YOLOv8 model to perform color difference analysis on the processed image, the multi-scale feature enhanced YOLOv8 model uses HDConv formed by combining convolution and hole convolution to extract features, and uses DSPPF combined with deep convolution and global pooling to perform multi-scale feature context enhancement; a result output module, used to display the detection results to users in a visual form, so that users can quickly understand and take corresponding processing measures.

[0005] Preferably, the image acquisition module uses a multispectral camera for acquisition. The multispectral camera divides the light into different bands based on the principle of spectral segmentation, and then acquires and processes them separately to show the color distribution and changes of the LED; the multispectral camera is used to shoot the target image under LED lighting conditions, and the shooting parameters of the multispectral camera are set according to the size, shape and position of the LED board: spectral band selection, exposure time, gain range, white balance, focal length, frame rate, shooting mode, color depth, and other parameters; since the color difference of the LED is mainly manifested in the spectral characteristics, the multispectral camera is used to capture the LED light of different spectral bands, so as to show the color distribution and changes of the LED; different LEDs are shot with the same shooting parameters to obtain different LED color difference images, and all images are synthesized into one data set.

[0006] Preferably, the preprocessing module includes an attention-guided denoising convolutional neural network ADNet to denoise the image so as to effectively remove the noise in the image while maintaining the image detail information; specifically, the data set synthesized by the image acquisition module is uploaded to a computer with a GPU, and the data set image of the image acquisition module is denoised by using an attention-based image denoising network ADNet (Attention-guided Denoising Net); the ADNet network can effectively remove the noise in the image while maintaining the image detail information through a combination of four modules: sparse block (SB), feature enhancement block (FEB), attention block (AB) and reconstruction block (RB); the data set of the image acquisition module is enhanced by rotation, scaling, translation and brightness increase to enrich the diversity of the data set image; the targets in the data set image of the image acquisition module are annotated using labelImg, including bounding box annotation and category annotation; the annotation information is saved in YOLO format.

[0007] Preferably, the detection and analysis module uses a multi-scale enhanced YOLOv8 model. The YOLOv8 color difference detection model further extracts image features by downsampling by using a self-developed convolutional layer instead of the original downsampling convolutional layer, and maps these features to feature maps of different scales, thereby increasing the model's ability to adapt to features of different scales. Specifically, the data set images of the image acquisition module are input as training samples into a neural network model of the YOLOv8 architecture for training to obtain a multi-scale enhanced YOLOv8 color difference detection model. The YOLOv8 color difference detection model uses a series of convolutional layers to further extract features. And map these features to feature maps of different scales; a series of convolutional layers of the YOLOv8 color difference detection model are: on the same feature, the feature is convolved differently, and the different convolutions refer to: first, ordinary convolution, second, convolution with different hole rates after ordinary convolution, and transformable hole convolution that fuses the results of convolutions with different hole rates by setting special weights; the different convolution results are connected and spliced, and the channels are reshuffled and fused. The output results are globally pooled to increase global contextual connections and reduce the amount of calculation. The convolution layer of the YOLOv8 color difference detection model is expressed as a hybrid dilated convolution (HDConv), and HDConv increases the model's ability to adapt to features of different scales; the YOLOv8 color difference detection model uses the self-developed DSPPF (Deep SpatialPyramid The YOLOv8 color difference detection model is a color difference detection model in which the output result of the convolution layer is used as the input feature for two-way linear transformation, and a q matrix is ​​generated for one of the paths; the other path is transformed for three layers of convolution, and finally the global maximum pooling is used to capture the global context of the entire input; the results of the three layers of convolution are residually connected, and the features are deeply fused; the fused features are dot-product multiplied with the q matrix, and the features extracted by the three layers of convolution are fused on the basis of the original input features, thereby enhancing the feature expression of the module; the YOLOv8 color difference detection model is connected to an enhanced feature network PAN, which is implemented by upsampling, splicing and convolution operations, so that the model can simultaneously utilize the feature information of low-level and high-level layers; the YOLOv8 color difference detection model is connected to a prediction head composed of a convolution layer and an output layer, and the output layer is responsible for generating the bounding box coordinates, object confidence and category probability, and the results are generated by decoding the output layer to achieve fast and accurate detection of color difference objects in the image;The YOLOv8 color difference detection model calculates the loss between the predicted result and the true annotation during the training process, uses the back propagation algorithm to calculate the gradient of the loss to the model parameters, and uses the SGD optimizer to update the parameters to minimize the loss and improve the model accuracy. ;

[0008] Preferably, the color difference detection system can output the position and category of color difference blocks of LED boards of different color types. The color difference detection system is efficient, accurate, flexible and scalable, and can significantly improve the production quality of LED boards and reduce production costs.

[0009] Preferably, the color difference detection system realizes user interaction through the Qt system interface. The user uses simple interface operations to realize the analysis of LED color difference and adjusts the production of LED boards in time according to the results to ensure that the quality meets the requirements.

[0010] Preferably, the product of the color difference detection system includes a computer program code stored on a computer-readable medium, and the product of the color difference detection system runs on a computer to implement the functions of the LED color difference detection system.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] The invention provides an LED color difference detection system based on computer vision. The system uses image processing technology to perform denoising and enhancement processing on LED board images, and extracts and analyzes features of LED board images through a deep learning algorithm based on computer vision, so as to realize efficient acquisition and accurate color difference analysis of LED board images. The system is composed of four core modules: an image acquisition module, a preprocessing module, a detection and analysis module and a result output module. The image acquisition module uses a multi-spectral camera to capture high-definition images of LED boards, thereby ensuring the richness and accuracy of data sources. The preprocessing module uses an attention-oriented denoising convolutional neural network ADNet to perform denoising, enhancement and other processing on images to improve detection accuracy. The detection and analysis module uses a multi-scale enhanced YOLOv8 target detection algorithm to perform color difference analysis on the processed images. The result output module displays the detection results to users in a visual form, so as to facilitate users to quickly understand and take corresponding processing measures. The system has the characteristics of high efficiency, accuracy, flexibility and scalability, solves the problems of low efficiency and high false detection rate of traditional LED color difference detection methods, and can significantly improve the production quality of LED boards and reduce production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is the overall flow chart of the present invention.

[0014] Figure 2This is a structural diagram of the multi-scale enhanced YOLOv8 color difference detection model provided by the present invention.

[0015] Figure 3 This is the S3 HDConv convolutional layer diagram provided by the present invention.

[0016] Figure 4 This is the S3 DSPPF diagram provided by the present invention.

[0017] Figure 5 This is a flow chart of the deployment of the user interface model provided by the present invention. DETAILED DESCRIPTION

[0018] In order to enable those skilled in the art to better understand the solution method of the present invention, the present invention is further described in detail below with reference to specific embodiments in conjunction with the accompanying drawings. However, this method is not limited to a specific example and can be applied to similar situations.

[0019] See also Figures 1 to 5 , using the LED color difference detection system based on computer vision provided by the present invention to perform color difference detection on LED:

[0020] Step 1: On the industrial production line, use a multispectral camera to collect images under LED lighting conditions under specific lighting conditions. Determine the selection of spectral bands according to the color of the LEDs produced by the industrial production line, green (G): 560nm±16nm, red (R): 650nm±16nm, red edge (RE): 730nm±16nm, near infrared (NIR): 860nm±26nm, and determine the band that can capture the spectral color information of the LED. Adjust the lens parameters according to the image's field of view, magnification, and light transmittance. The shutter speed determines the exposure time. According to the lighting conditions under LED lighting conditions, select the shutter speed: 1 / 30 to 1 / 12800 seconds. Adjust the gain to adjust the brightness of the image, 1x to 32x, to adapt to different lighting conditions. White balance, focal length, frame rate, shooting mode, color depth, and other parameters are the same as ordinary cameras. Fine-tune according to actual conditions and select the most appropriate parameters. Multiple groups of LED images are taken with the same parameters. Due to the mass production of industrial production lines, the LED images taken are within 10 colors. In addition, LED images of multiple angles are taken in a single batch of a single color. When LED images of different batches of a single color are taken, the number of images must be at least thousands to form a data set.

[0021] Step 2: Upload the dataset to a computer terminal with a GPU, preprocess the dataset images, and use the ADNet network to denoise the images, removing noise while maintaining image detail information. Use LabelImg to mark the LED areas with obvious color difference in the image, with the marking information being the position and the label information being the color difference. Then, data enhancement is performed by rotating, scaling, translating, and adjusting brightness to increase the richness and diversity of the dataset.

[0022] Step 3: Input the preprocessed dataset into the multi-scale enhanced YOLOv8 model for training. The model uses HDConv with different void rates of 1, 2, and 3 for downsampling feature extraction. HDConv is placed in the backbone network for downsampling to extract features of different scales. The model uses DSPPF to replace the original SPPF. The deep convolution layer in DSPPF replaces the maximum pooling layer in the original SPPF to capture the contextual information of features of different scales, and the feature expression of the model is enhanced through residual connection and maximum pooling. Finally, the bounding box position, category, and confidence are generated through the neck network and the detection head output to achieve fast and accurate detection of color difference objects in the image.

[0023] Step 4: Use Qt (C++ version) to set up the user interaction interface, select the captured image and upload it to the interface, call the trained model to detect LED color difference, output the color difference category and color difference position, and display them in the output result column. Users can select defective LED boards based on the output results for subsequent optimization and improvement.

[0024] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the above examples. Any changes, modifications, additions or substitutions made by those skilled in the art within the essential scope of the present invention shall also fall within the protection scope of the present invention. The present invention does not elaborate on some of the known technologies of those skilled in the art.

Claims

1. A LED color difference detection system based on computer vision, characterized in that: The system includes: Image acquisition module, used to obtain LED board color difference image data set; The preprocessing module is used to perform denoising and enhancement on the LED board color difference dataset images to improve the detection accuracy; The detection and analysis module uses a multi-scale feature enhanced YOLOv8 model to perform color difference analysis on the processed image. The multi-scale feature enhanced YOLOv8 model uses HDConv formed by combining convolution and dilated convolution to extract features, and uses DSPPF combined with deep convolution and global pooling to perform multi-scale feature context enhancement. The result output module is used to display the test results to the user in a visual form, so that the user can quickly understand and take corresponding treatment measures.

2. The LED color difference detection system based on computer vision according to claim 1, characterized in that: The image acquisition module uses a multi-spectral camera to collect light. Based on the principle of spectrum segmentation, the light is segmented according to different wavelengths. Its optical system collects and processes light to show the color distribution and changes of the LED.

3. The LED color difference detection system based on computer vision according to claim 2, characterized in that: The preprocessing module includes an attention-guided denoising convolutional neural network ADNet to denoise the image so as to effectively remove the noise in the image while maintaining the image detail information.

4. The LED color difference detection system based on computer vision according to claim 3, characterized in that: The detection and analysis module uses a multi-scale enhanced YOLOv8 model. The YOLOv8 model further extracts image features by downsampling by using a self-developed convolutional layer instead of the original downsampling convolutional layer, and maps these features to feature maps of different scales, thereby increasing the model's ability to adapt to features of different scales.

5. The LED color difference detection system based on computer vision according to claim 4, characterized in that: The multi-scale enhanced YOLOv8 model includes a deep convolutional layer instead of the maximum pooling layer stack in the SPPF in the unenhanced model to capture the contextual information of features of different scales, and then performs residual connection after global maximum pooling, and finally performs dot product multiplication to perform feature fusion.

6. The LED color difference detection system based on computer vision according to claim 5, characterized in that: The color difference detection system can output the position and category of color difference blocks of LED boards of different color types.

7. The LED color difference detection system based on computer vision according to claim 6, characterized in that: The color difference detection system realizes user interaction through the Qt system interface. Users use simple interface operations to analyze the LED color difference and make timely adjustments to the production of LED boards based on the results.

8. The LED color difference detection system based on computer vision according to claim 7, characterized in that: The product of the color difference detection system includes a computer program code stored on a computer-readable medium. The product of the color difference detection system runs on a computer to implement the functions of the LED color difference detection system.

Images