Image semantic segmentation with an improved fully convolutional network

Abstract

With the development of deep learning and the emergence of unmanned driving, fully convolutional networks are a feasible and effective for image semantic segmentation. DeepLab is an algorithm based on the fully convolutional networks. However, DeepLab algorithm still has room for improvement, and we design three improved methods: (1) the global context structure module, (2) highly efficient decoder module, and (3) multi-scale feature fusion module. The experimental results show that the three improved methods that we proposed in this paper can make the model obtain more expressive features and improve the accuracy of the algorithm. At the same time, we do some experiments on the Cityscapes dataset to further verify robustness and effectiveness of the improved algorithm. Finally, the improved algorithm is applied to the actual scene and has certain practical value.