Image Recognition And Defect Detection Based On Deep Learning

The current era is the era of artificial intelligence.How to make computers intelligent and capable of autonomous learning like the human brain is a hot area of concern for many researchers.Deep learning,as the core method of artificial intelligence,has a very wide range of applications in this field.In fact,deep learning involves all aspects of life,including but not limited to image recognition,defect detection,human-computer interaction,unmanned driving,and big data analysis.This article will use deep learning methods to research some image processing issues,which can be divided into two parts,namely using image recognition method based on deep learning for facial expression recognition,and using target detection technology based on deep learning to achieve defect detection of industrial gears.The main contents of this article are:1.For the facial expression recognition,using face detection technology,LBP feature extraction,size normalization to preprocess the images in the facial expression database,making the images more suitable as input for neural networks.2.Using the improved LeNet-5 network model to recognize facial expressions.The original LeNet-5 network did not perform well on facial expression recognition.Therefore,the LeNet-5 network was improved,including the changes of network parameters,and adjustments of the network structure.The experimental results show that,compared with the original algorithm,the accuracy of the algorithm in this paper on the CK + and JAFFE databases has been improved by 12.88% and 13.68%,respectively.3.The improved network structure is tested on the CK + and JAFFE data sets,and the accuracy rate is obtained,then compare it with the original network structure and other algorithms.4.Use the improved YOLOv3 network to detect gear defects.The improvement measures are to use the Dense Net network instead of the original Darknet-53 network and adjust the prediction scale.5.Construct a gear defect images data set,including image acquisition and expansion and defect labeling,and then using the improved YOLOv3 network to test and evaluate on the data set,including network structure performance evaluation,algorithm performance comparison,and the comparison of algorithm performance under different light intensities.Experimental results show that,compared with the original algorithm,the average accuracy of the algorithm used in this paper is improved by 3.87%,and the accuracy of the missing parts of the gear is increased by 5.7%...