WCSE

An Improved End-to-end Method for Improving Smoked Image Recognition in Outdoor Rescue Scene

Abstract— This paper discusses a method for image clarity during rescue vehicle operations. This method first deals with the uncertainty caused by image jitter, and then uses defogging methods to improve the image, bringing more convenience and safety to unmanned rescue processes. The dehazing process is processed using an improved AOD Net method. Firstly, replace ReLU with LeakyReLU to minimize peripheral interference, and preprocess the collected images for brightness before sending them to the detector. Secondly, using bilateral mesh filtering to make the dark scene clearer after defogging. Through experiments, it can be found that the method can improve atomized images and maintain the brightness of the images, which can help to enhance the visual experience.

Index Terms— DOA-Net, Smoke detect, Dehaze Image, image enhancement, color recovery.

Lei Zhang
Shanghai Fire Research Institute of MEM, CHINA
Zhen Ruan
Shanghai Fire Research Institute of MEM, CHINA
Zhen Li
Shanghai Fire Research Institute of MEM, CHINA
Xi Liu
Shanghai Fire Research Institute of MEM, CHINA
Hongliang Chen
Shanghai Fire Research Institute of MEM, CHINA
Ziyou Wang
Shanghai Fire Research Institute of MEM, CHINA

[Download]

Cite: Lei Zhang, Zhen Ruan, Zhen Li, Xi Liu, Hongliang Chen, Ziyou Wang, "An Improved End-to-end Method for Improving Smoked Image Recognition in Outdoor Rescue Scene," 2024 The 14th International Workshop on Computer Science and Engineering (WCSE 2024), pp. 58-62, Phuket Island, Thailand, June 19-21, 2024.