Improving YOLO-based Object Detection with A Universal Restoration Network in Complex Environments

Object detection based on deep learning has been widely applied in the field and has achieved many achievements, including improving accuracy, real-time performance, and expanding application scope. However, when the environment becomes complex, such as in rainy environments, camera shake leading to blurring, etc., low object detection rate and easily-missed detection problems may occur. This article proposes a YOLO-based object detection scheme to address this issue. The main idea is to restore high-resolution images from their degraded versions before sending them into the network of YOLOv5. To this end, a powerful and universal image restoration network is integrated into the solution to deal with various types of recovery tasks, such as image de-raining, deblurring, and denoising. Experimental results show that the integrated scheme performs better in object detection than the original YOLOv5 algorithm, leading it to a practical application in thepresence of complex interference.