After all these years of rapid development, as the economic level of our country has risen, new challenges have arisen for productive work in a polluted environment. Due to hazy weather, the quality of images captured by traffic monitoring devices could be degraded, affecting the accuracy of judging vehicle dynamics. In this background, we introduce MSTBNet, a fresh network design utilizing multi-scale topological residual blocks (MSTBs) to convey information across both depth and width dimensions while effectively identifying network nodes to enhance learning. The network can capture both local and global features through the transfer of information between neighbouring and distant nodes and the feedback mechanism of the network implementation. Also, due to the lightweight of the network model, it is well suited for real-time operation of target detection. Experimental results show that our proposed algorithm has superior dehazing performance compared to existing methods, which improves the accuracy of target detection and identification.