Research On Aluminum Surface Defect Detection Based On Machine Vision

Surface defect detection technology,as an essential part in the product quality control process,can effectively improve aluminium product quality and production efficiency.Traditional machine vision detection methods rely on the design of feature extractors.The performance of the feature extractor determines the performance of the detection method.However,the artificially designed feature extractor's extraction capability is relatively single,and it is easy to miss many valuable features.As a particular branch of machine learning,deep learning can automatically learn parts and can detect multi-directional information about the target at the same time to perform qualitative and quantitative analysis of defects.However,the problems in cluding small data sets are common in industrial defect detection,which hinders the development of deep learning technology in the field of aluminum surface defect detection.The use of deep learning methods to study the surface defect detection of aluminium materials is of great significance to improve aluminium materials' production efficiency and improve the surface quality of aluminium products.This theis research into the following four aspects:(1)An auxiliary annotation algorithm for image segmentation is proposed.First,the superpixel algorithm is used to split the original image into a combination of multiple superpixels to obtain the rough outline of the defect area;secondly,each superpixel is aggregated by the density clustering algorithm to get a complete and clear defect segmentation boundary;and finally,For the segmentation boundary,the corner detection algorithm is used to obtain the boundary intersection point as the recommended point of the label.Simultaneously,to optimize the problem that the algorithm cannot be segmented when the difference in image colour distribution is slight,an adaptive threshold algorithm is designed according to the overall colour distribution of the image,which effectively avoids this problem.Significantly improve the efficiency of labelling.(2)An improved segmentation decision network model is proposed to detect surface defects of aluminum materials.On the basis of the original segmentation decision network,an aluminum area segmentation network is added between the input layer and the output layer of the segmentation network to segment the aluminum area,avoiding the interference of the aluminum background on the detection results,and using multi-task training In this way,the generalization of the network is improved;after segmenting the network output layer,the SE(Squeeze-and-Excitation)module is introduced,and the attention mechanism is added,so that the network can better utilize the advantages of multi-task feature fusion and improve the network's ability to extract important features.Through experiments,the collocation of the loss function of the segmented network in this article is determined.The segmentation network uses the cross-entropy loss function as the loss function,and the classification network uses the root mean square error loss function as the loss function.(3)An experimental platform is built to verify and analyze the improved segmentation decision network model proposed in this paper.Based on the method proposed in this article,select the required hardware and software environment and appropriate evaluation indicators.Two sets of experiments are designed to verify the effectiveness of the improved strategy verification experiment and the comparison experiment with other methods to verify the effectiveness of the method proposed in this paper.Record experimental data and visualize it.Experiments have proved that the method proposed in this paper can basically solve the problem of surface defect detection of aluminum materials.