Research on Acceleration Methods of Semi-training Color Stripping DehazeNet

This paper discussed the influence of the network size of the color stripping module on the dehazing performance of STCSDN through analysis and studies of the color stripping module structure in semi training color stripping DehazeNet (STCSDN). After a large number of tests, this paper verified that the performance of simplified network is close to the original network in image dehazing by removing redundant parameters and simplifying the model structure. Furthermore, the simplified network processed images faster, made the network can conduct real-time image dehazing tasks, and improved the practicality of the network.