IIE-SegNet: Deep Semantic Segmentation Network With Enhanced Boundary Based on Image Information Entropy

Abstract

With the vigorous development of deep learning and the widespread use of mobile robots, automatic driving has gradually become a research hotspot. Environment perception is the most important part of automatic driving technology, and the purpose of environment perception is to distinguish the environmental content. Therefore, accurate and efficient image semantic segmentation method is becoming more and more important. In this paper, we introduce a deep semantic segmentation solution: IIE-SegNet: Deep semantic segmentation network with enhanced boundary based on image information entropy. At present, deep learning based on semantic segmentation solutions has some problems, such as low segmentation accuracy for small-scale objects and unclear boundary of segmented objects. Our method preserves the boundary of the segmentation object, and has higher segmentation accuracy for small-scale objects. In our method, the features of the underlying pooling layer are added to the ASPP structure of the encoding module, and the image information entropy of the previous pooling layers is introduced into the decoding module. We also introduce focal loss to solve the problem of imbalance between positive and negative samples. Finally, the test results on the extended Pascal VOC 2012 test set, abbreviated to Exp-Pascal VOC 2012 test set show that the proposed method has better performance on Exp-Pascal VOC 2012 test set compared with the advanced methods at the present stage, the segmentation accuracy of small-scale targets is higher, and the boundary is clearer.

Keywords

• Semantic segmentation
• deep learning
• enhanced boundary
• image information entropy