Research On Online Visual Detection Method Of Aluminum Profile Surface Defects Based On Deep Learning

In recent years,the development scale of aluminum profile industry is further expanding.Different kinds of defects often appear on the surface of aluminum profiles in actual production,timely and effective detection of its surface defects is of great significance for later reuse.At present,the detection of the surface defects of aluminum profiles is still based on the manual visual detection method,which cannot meet the needs of industrial mass production because of its limitations.With the development of industrial automation towards the direction of intelligence,some methods based on digital image processing have been applied in metal surface defect detection.However,this method often requires researchers to design corresponding features according to different defect categories.Obviously,such feature construction method lacks robustness and versatility.With the development of deep learning technology,metal surface defect detection methods based on deep learning have been widely used.In this paper,an online visual detection method for surface defects of aluminum profiles based on deep learning is proposed.The main contents are as follows:Aiming at the problem of lack of data sets on surface defects of aluminum profiles.Firstly,based on the surface defect data set of aluminum profiles published by Aliyun,this paper analyzes the defect categories and characteristics of the data set.Then perform data enhancement operations including image flipping,rotation,multi-scale scaling,gamma change,etc.The purpose is to expand the amount of data and enrich the diversity of data.Finally,the expanded data set is marked and divided to obtain a standard aluminum profile surface defect data set.A method for surface defect detection of aluminum profiles based on Yolov3 network is proposed.Through the analysis of experimental results,it is found that the detection of aluminum profile surface defects based on Yolov3 network has a high rate of missed detection and false detection.Therefore,this paper proposes a surface defect detection method for aluminum profiles based on Improved Yolov3 network,including:According to the actual size distribution of the surface defects of the aluminum profile,the size of the anchor box was redesigned using K-means algorithm.In order to improve the detection accuracy of small target defects,the network structure of the multi-scale prediction part of Yolov3 network was modified.In order to get a more accurate prediction of the boundary box,the GIOU loss is used as the regression loss function.The experimental results show that the detection effect of the Improved Yolov3 network is better than that of the Yolov3 network.The m AP value is increased from 84.12% of the Yolov3 network to 90.6%,an increase of 6.48%,and the detection time of a single image reaches 37.6ms.Although the detection effect of Improved Yblov3 network is better than that of Yolov3 network,it still has the problem of poor detection effect of small target defects such as paint blisters and dirty spots.In order to further improve the detection effect of small target defects,an aluminum profile surface defect detection method based on Faster R-CNN network is proposed,which includes:Create a defect data set containing only two types of paint bubbles and dirty spots,and redesign the size and aspect ratio of the anchor based on this data set.ROI-Align layer is used to avoid the influence of pixel deviation on the regression location of small target defects.Facing the problem of poor detection effect caused by dense defects of small targets,the Soft-NMS algorithm is used to eliminate redundant bounding boxes.The experimental results show that the AP value for paint bubble defects is increased from 51.07%to 64.06%,and the AP value for dirty spot defects is increased from 77.39% to 83.98%,and the single image detection time is about 100 ms.Finally,the construction and application of the surface defect detection platform for aluminum profiles are realized.Choose suitable industrial cameras,lenses,light sources,lighting methods,and supporting structure components to build a defect detection platform.Then apply the deep learning model to the platform,and artificially mark certain types of defects to realize real-time collection of surface images of aluminum profiles and defect detection to verify the effectiveness and feasibility of the designed model.