Multi-Scale Enhancement and Sharpening Method for Visible Light Images in Underground Coal Mines

The dusty and fog may cause inefficient production and safety hazard for staffs on the underground of coal mine. According to this characteristic, the visible image sharpening method is studied in our work. In fact, some known methods cannot fit on the underground of coal mine environment. Moreover, model complexity needs to be reduced to help real-time processing. Hereby, we provide a sharpening method based on multi-scale enhancement in this paper. The end-to-end Gated Context Aggregation Network (GCANet) is the baseline of the proposed one. Moreover, the multi-scale feature enhancement strategy is the key point to capture and fuse different level information. To reduce overall computational complexity of the model, a lightweight octave convolution is used to replace the classical convolution operation. In experiment part, the public dataset, i.e., RESIDE, and the measured underground coal mine dataset, i.e., Mine-dehaze, are sharpened by the proposed method. In addition, the sharpening result is compared with ones by some known methods, which can reveal the superiority of the proposed one.