Abstract：To address the problem that deep convolutional encoder-decoder network do not take into account the channel dependence on each convolutional feature map in road scene segmentation,a deep convolutional encoder-decoder network that incorporates the channel attention mechanism was proposed,and the channel attention mechanism was improved to a dual channel attention mechanism.The method retained the advantage that the original channel attention mechanism can optimize background information,while adding another channel for collecting important features between the difficult to distinguish objects to obtain detailed channel attention.For the road scene images,the experimental results show that the deep convolutional encoder-decoder network with the dual channel attention mechanism further improves the segmentation performance,which improves the evaluation accuracy and intersection over union by about 7 and 8 percentage points respectively.
黄泽华,丁学明. 融合通道注意力机制和深度编解码卷积网络的道路场景分割[J]. 小型微型计算机系统, 2021, 42(11): 2362-2367.
HUANG Ze-hua,DING Xue-ming. Road Scene Segmentation Based on Deep Convolutional Encoder-decoder Network with Channel Attention Mechanism. Journal of Chinese Computer Systems, 2021, 42(11): 2362-2367.
2021, Vol. 42 
