Abstract: Due to the existence of a large number of smoke in the fire scene, the image clarity of the video monitoring system becomes blurred, and the contrast and clarity of the image decline, which can not provide effective visual support for evacuation and search and rescue. In view of this situation, this paper proposes a gray-scale image smoke removal algorithm based on deep learning. The network proposed in this paper is mainly composed of two parts: detection sub network and removal sub network. The former determines the specific location of smoke through residual learning network, and the latter uses dense U-shaped network to remove smoke while retaining the original background, and uses dense block to reuse low-level features to high-level features to further improve the accuracy of smoke removal. A large number of experimental results show that the network has better performances in removal effect and real-time, and the subjective evaluation and objective evaluation are better than other comparison algorithms.

Key words: deep learning, image desmoking, fire rescue, gray-scale image, video monitoring