Research On Semantic Segmentation Of Road Scene Based On Deep Neural Networks

Image semantic segmentation is a key and difficult task in the field of computer vision.It is widely used in practice.The semantic segmentation of road scenes is one of the core technologies of autonomous vehicles.Therefore,the study of image semantic segmentation technology has important practical significance.Moreover,with the computer vision stepping into the era of deep learning,the difficulties of image semantic segmentation are gradually being overcome.Researchers have successively proposed a series of semantic segmentation methods based on the convolutional neural networks,and frequently refresh the segmentation accuracy.First of all,this thesis summarizes the research status of image semantic segmentation and analyzes the research results of advanced neural networks used in semantic segmentation related fields.Secondly,an ERFNet-Efficient network structure based on a deep convolutional neural networks is designed.The network structure is an improvement over the encoder-decoder structure ERFNet network.By using the latest research results such as asymmetric convolution,group convolution,dilated convolution,transposed convolution,batch normalization,the semantic segmentation task of the road scene can be accurately and efficiently completed.Thirdly,the ERFNet-Efficient model was implemented using the PyTorch deep learning framework.Then training and testing were performed on the Cityscapes dataset,and good results were obtained.Fourthly,the segmentation accuracy of the ERFNet-Efficient model has been further improved through related technologies such as transfer learning and ensemble learning.Fifthly,the ERFNet-Efficient network model of this thesis is compared with the ERFNet algorithm and other the most advanced algorithms.The performance indicators are model accuracy,operating speed,and model parameters.The results show that ERFNet-Efficient network is one of the most excellent algorithms in overall performance.