Research On Image Multi-classification Method Based On Convolutional Neural Network

With the development of technology,the application of images in various fields has become more extensive,especially the image multi-classification technology.Image multi-classification techniques include image edge detection,object recognition,semantic segmentation.At present,many researcher at home and abroad are devoted to the research of multi-classification technology of images.The principle of image multi-classification method has two main directions,one is image multi-classification method using non-convolution neural network;the other is image multi-classification method based on convolutional neural network.This paper introduces three kinds of image classification methods in the above two types.The image multi-classification methods of non-convolution neural network are Linelet,CGVS,DGED and image multi-classification method based on convolutional neural network.HED,Faster-RCNN,Deeplab-V3+.This paper mainly summarizes the above algorithms,realizes the research of image multi-classification method based on convolutional neural network,and compares and analyzes the advantages of image multi-classification method based on convolutional neural network in image multi-classification.This paper first expounds the theoretical principles of the above six methods,and introduces two data sets BSDS 500 and PASCAL VOC 2012 which are commonly used in the field of image multi-classification.The image multi-classification method of non-convolution neural network and the image multi-classification method based on convolutional neural network are simulated.The test set on BSDS 500 and the test set on PASCAL VOC 2012 are tested respectively.Simulation test,comparative analysis of the results of the six methods of Linelet,CGVS,DGED,HED,Faster-RCNN,Deeplab-V3+.The analysis and summary of the image multi-classification method based on convolutional neural network has the advantages of high detection accuracy,fast running speed and stable detection results.It is precisely because of these advantages that image multi-classification methods based on convolutional neural networks become more and more popular.Applications are becoming more widespread.