基于注意力机制和扩张解码改进的语义分割研究

Research on semantic segmentation based on attention mechanism and extended decoding

随着自主系统的兴起,技术革新和生产生活中需求的不断加大,实时性计算越来越受欢迎。在本文中介绍了快速卷积神经网络(Fast-SCNN),这是一种基于高分辨率图像数据的实时语义分割模型。在该模型现有的2个快速分割分支基础上,将原网络解码器部分改进为扩张解码,扩大感受野,有效提升网络模型的分割精度。然后引入了CBAM注意力机制模块,减少对冗余信息的关注,降低了计算量,提高了分割效率。改进后的网络在Cityscapes数据集上获得了73.27%的MIoU,同时保证了网络的推理速度,实验结果表明改进网络较原网络性能有所提升。

With the rise of autonomous systems,technological innovation and increasing demand in production and life,real-time computation is increasingly desirable.This paper introduces fast segmentation convolutional neural network(Fast-SCNN),a real-time semantic segmentation model on high resolution images data.Building on existing two-branch methods for fast segmentation,the paper improves the original network decoder to the extended decoding,expands the receptive field,and effectively improves the segmentation accuracy of the network model.Then the attention mechanism module of CBAM is introduced to reduce the attention to redundant information,reduce the amount of calculation and improve the segmentation efficiency.The improved network obtains MIoU of 73.27%on Cityscapes data set,and ensures the inference speed of the network.The experimental results show that the performance of the improved network is improved compared with the original network.