SimpleISP-YOLO: A Lightweight Model for Low-Light Image Enhancement and Object Detection by Simulating Simplified ISP Process

In recent years, the field of object detection has seen the rise of numerous remarkable models due to the advancement of deep learning technologies. Low-light conditions have historically posed a significant challenge in the domain of object detection. The quality of the initial imagery plays a pivotal role in determining the efficacy of subsequent object detection algorithms. Notably, in dimly lit settings, the captured images are prone to heightened noise, color aberrations, and various other challenges. To address these issues, the current study presents a novel low-light image enhancement framework, designated as SimpleISP-Net. Inspired by Image Signal Processing (ISP), SimpleISP-Net simulates the simplified ISP process by performing only color correction and gamma correction on low-light images to achieve image enhancement. By integrating SimpleISP-Net with the YOLOv5s detection model, an end-to-end low-light object detection model called SimpleISP-YOLO is then constructed. The experimental results show that the algorithm proposed in this paper achieved an mAP of 68.1% on the low-light target detection dataset ExDark, which is an improvement of 11.4% compared to the original YOLOv5s model, with only a slight increase in computational complexity by 2.83 GFLOPs.