Underwater Object Detection Based on Enhanced YOLO
Xiaohan Wang, Zetao Jiang, Zhaoqiang Xia, Xiaoyi Feng
Abstract
As an important research topic in the field of computer vision, object detection has been successfully applied to several fields. YOLO is one of the popular frameworks for detection, but the traditional YOLO detection method lacks the processing of anchor points with detection and recognition features. In addition, most detection methods seldom consider of complex environments, especially for underwater images with high turbidity. Therefore, a YOLO based underwater object detection method for underwater images is proposed. An improved YOLO detection method without anchor points is introduced, where the detection features are separated from the recognition features to reduce the mutual interference between features and improve the detection accuracy. Further, a Retinex-based image enhancement algorithm is also proposed for underwater images enhancement. Relevant experiments based on underwater datasets are conducted to verify the effectiveness of the proposed enhanced YOLO detection method.