Image Semantic Segmentation Method Based On Multi-scale Cascade Network

Semantic image segmentation aims to classify each pixel of the images based on the different semantic meanings according to the input images.This paper focuses on how to improve the computational efficiency of the algorithm as much as possible while ensuring the accuracy of image segmentation.Apart from that,this paper proposes a multi-scale cascade network.The related works are as followsFirstly,for complex image semantic segmentation problems,this paper mentions the MainNet which fuses/combines the atrous convolution algorithm and the pyramid pooling algorithm.And the feature fusion improves the network's ability to understand complex scenes.Secondly,aiming at the problem of long computation time for current deep neural net-work image semantic segmentation algorithm,a fast image semantic segmentation system based on multi-scale cascade network is proposed,which effectively solves the problem of low computational efficiency for image semantic segmentation based on deep neural net-work.A multi-scale cascaded network is constructed consisting of encoder-decoder archi-tectures which can achieve incorporating image scaling elements into the network,providing low-resolution images with semantic information and high-resolution images with detailed information.Finally,this paper shows that the proposed method leads the classical algorithm greatly in computational efficiency based on the comparison experiments with classical algorithms on image segmentation datasets such as Cityspaces,ADE20K,etc...