Research On Visual On-line Detection Technology Of Laser Paint Removal Effect

Laser paint removal is an important application in the field of laser cleaning.However,in the field of industrial cleaning,paint removal detection relies on manual visual inspection,and the lack of effective online detection methods limits the wide application of laser paint removal.In the laser paint removal process,the color tone,saturation,metal reflectivity and texture of the cleaning target change with the paint removal process.These image information contains a lot of potential information,and machine vision detection technology can be used to realize the paint removal process.online detection.In order to realize the visual online detection of laser paint removal effect,this paper uses the algorithm framework of deep learning to study the visual inspection model of laser paint removal effect,and builds a laser paint removal visual inspection experimental platform for the cleaning process and visual inspection technical problems of laser paint removal.And online visual inspection software,can realize real-time inspection of laser paint removal effect.The main research contents are as follows:(1)The superiority of detection technology based on machine vision for laser paint removal detection is discussed.For the nonlinear physical change process of laser paint removal,using the learning ability of deep learning for detection modeling has obvious advantages.According to the visual inspection requirements of paint removal effect,a visual experiment platform for laser paint removal is built for image detection research,which can provide real-time,stable and high-quality image data of paint removal effect for visual inspection of paint removal.(2)In order to construct the image data set necessary for the deep learning framework and realize the conversion of image prior knowledge,the experiment explores the best method of collecting image data for paint removal from the aspects of sample preparation,laser paint removal process research,and paint removal visual experiment.Provide reliable and rich image discrimination information for paint removal effect detection,establish the discrimination standard for paint removal visual detection,complete the image annotation of the corresponding paint removal effect,and establish a classification and segmentation laser paint removal image data set,so as to study and analyze the characteristics of paint removal image detection and its correlation.The data problems of the paint removal data set,and the technical problems of visual detection of paint removal effects are summarized.(3)The experiment uses the deep learning algorithm model to study the visual detection technology of paint removal effect,and constructs the detection model from the backbone network and the functional network.Through the comparison of the performance of multiple convolutional backbone networks and the study of strategies,the experiment proves the effectiveness of the convolutional neural network for the classification and detection of laser paint removal,and establishes the design route of using the Resnet lightweight backbone network.On the functional network,the compact segmentation network structure design completed by the Bi Se Net architecture is improved,and the loss function is designed for the noise samples to balance the problem of difficult and easy sample learning,overcoming the inaccurate contour detection of the segmented image and the noise learning problem,and the segmentation effect reaches 0.9277 MIo U,completes the semantic description of the painted metal surface.In order to reduce the performance requirements of computing equipment,the algorithm model optimization was completed in the experiment,and the low-power CPU reached a real-time detection speed of 11.4FPS,which has practical application value.(4)In response to the detection requirements of the paint removal effect of the experimental samples,the laser paint removal visual online detection software was built,and the software detection structure,detection function development and GUI interface design were completed,which can realize the positioning of the incoming sample and the segmentation detection of the cleaning process.,the state after cleaning and other functions,the detection software can effectively detect and feedback the paint removal effect in the laser cleaning process to achieve the purpose of visual online detection.