Abstract: To address the problems of too many parameters and much time consumption in existing recognition networks for transmission lines, a lightweight encoder-decoder network is constructed to discern complex transmission line images featured with multiple intersections quickly and accurately. The encoder is based on the first 16 layers of conventional MobileNetV3 to reduce network parameters. The convolutional block attention module is used to replace the squeeze and excitation attention module to improve the network's ability to extract the feature information of transmission lines. The decoder is constructed by combining deeply separable convolution and deep atrous spatial pyramid pooling to expand the receptive field and improve the network's ability to aggregate contextual information with different scales. Moreover, the training network is sparse by using the L1 regularization method. The pruning threshold is determined according to the product of the scaling factor and the corresponding output of each channel to remove redundant channels and compress the network effectively, which improves the recognition speed of transmission lines. Experimental results demonstrate that the mean pixel accuracy, mean intersection over union , and recognition speed of the lightweight encoder-decoder network are 92.11%, 84.19%, and 41 frames per second, respectively, which are better than PSPNet, U2Net, and existing improved transmission lines recognition networks.