Lightweight semantic segmentation algorithm based on MobileNetV3 network

With the popularization of intelligent terminals, more and more image segmentation tasks need to be carried out on mobile terminals. However, due to the limitation of mobile terminal computing and storage capacity, many classical semantic segmentation network models are too large to be applicable to mobile terminals. Therefore, this paper proposes a lightweight semantic segmentation algorithm based on the improved MobileNetV3 network. First, the leaky-ReLU6 activation function presented in this article was applied to the first seven layers of the MobileN etV3 network, and then the improved MobileNetV3 was used as an encoder for feature extraction. Then, the decoder part uses bilinear interpolation and depth separable convolution. Finally, the jump connection structure is retained to fuse the features of the lower layer and the higher layer. By comparing the network proposed in this paper with classic networks such as FCN, SegNet, UNet and DeepLabV3+, it is found that the network designed in this paper can effectively guarantee the segmentation accuracy, greatly reduce the number of parameters and shorten the running time of the network. It can be better applied to the mobile terminal with tight computing resources and space resources.