Semantic Segmentation Algorithm Based On Fully Convolutional Neural Network Model

Semantic segmentation is a research emphasis in the field of computer vision and pattern recognition,Traditional deep convolution neural network model based on image block image block period,assuming all pixels share the same label its sensitive to image spectrum and texture feature transformation,semantic segmentation result of target edge pixels segmentation accuracy is not high.To solve above problems,this paper puts forward a whole convolution neural network model based on residual module and multi-scale hollow convolution model.First of all,the full convolutional neural network model based on the residual module constructs the full convolutional neural network that can realize "end-to-end" training based on the residual module,and USES jump connection to improve the transfer efficiency of low-level detail features,so as to improve the image semantic segmentation accuracy.Secondly,on the basis of the full convolutional neural network model based on the residual module,the multi-scale cavity convolution model USES the cavity convolution to learn the different scale features of the original image,and combines these features to achieve the purpose of simultaneously learning the target detail features and global features.Finally,to verify the validity of the proposed algorithm,the proposed algorithm is applied to the ISPRS Vaihingen data set to compare and analyze the image block model.The experimental results show that the full convolutional neural network model and the multi-scale cavity convolution model can better learn the edge details of the image target.Compared with the traditional image block-based neural network and the improved Seg Net network model,the semantic segmentation accuracy can reach 84.56% and86.59%.The paper has 55 pictures,5 tables,and 50 references.