Defect Detection Of Aluminum Surface Based On Deep Convolutional Neural Network

With the in-depth application of information technology in the field of industrial manufacturing,big data research in industrial manufacturing is becoming an important reference for realizing intelligent manufacturing and helping the government guide the transformation and upgrading of manufacturing enterprises.In the traditional metal manufacturing industry,such as steel and aluminium,there are some problems,such as extensive production mode and simple production process.Therefore,it is urgent to improve production process and production efficiency by using the new generation of information technology such as artificial intelligence.Aluminum manufacturers should monitor the quality of aluminum materials before they leave the factory.The most important thing is to detect the surface defects of aluminum materials.On the other hand,aluminum surface defects are of many types,different sizes,and complex shapes.Human eyes usually cannot accurately identify surface defects in time.However,traditional machine vision detection methods are slow and accurate because they are manual feature detection.On the other hand,smaller defects are difficult to detect.With the vigorous development of computer networks,this paper mainly uses the deep convolutional neural network method to detect surface defects of aluminum.Firstly,the traditional detection algorithms are slow and the accuracy is not high.This paper mainly uses two mainstream object detection algorithms based on deep convolutional neural networks to detect the aluminum defect data sets,namely Faster R-CNN and YOLOv3.The former continues to use the design of the heuristic region proposals,but the overall process of the algorithm is based on a convolutional neural network.This design makes the effectiveness and efficiency of defect detection improved.The latter also uses the design of region proposals,but the overall design of the network is end-to-end,that is,the input to output is a unified whole,which makes it superior in detection efficiency.Finally,for the small defects on the aluminum surface,an improved algorithm based on YOLOv3 is proposed.For the design principle of the anchor box and principle of receptive field,we add an anchor box on the feature map of each layer to make the aluminum The detection effect of small defects on the surface of wood has been improved.Experiments were performed on the“Aluminum Defect Recognition”data set provided by the Guangdong Industrial Intelligent Big Data Innovation Contest.The experimental results show that the average accuracy(mAP)of the improved algorithm is 3.4%higher than YOLOv3 and 1.8%higher than Faster R-CNN.