Improved Lightweight Semantic Segmentation Algorithm Based on DeepLabv3+ Network

Due to the large number of semantic segmentation model parameters and time- consuming algorithm in deep learning, it is not suitable for deployment to mobile terminal. To solve this problem, a lightweight semantic segmentation algorithm based on improved DeepLabv3+ network is proposed. First, MobileNetv3 is used to replace the original DeepLabv3+ semantic segmentation model backbone network for feature extraction to reduce the complexity of the model and speed up the running speed of the model; second, the standard convolution in atrous spatial pyramid pooling module is replaced by depthwise separable convolution to improve the efficiency of model training; finally, the attention mechanism module and group normalization method are introduced to improve the segmentation accuracy. The proposed segmentation algorithm achieves a mean intersection over union (mIoU) of 72. 94% on the Cityscapes validation set of semantic segmentation dataset. Experimental results show that compared with common segmentation algorithms such as SegNet, Fast-SCNN, and ENet, the proposed algorithm can improve the segmentation effect while reducing the number of model parameters.