Improved YOLOv8-based video supervision technology for ulva prolifera salvage safety operations

Under the background of frequent occurrence of ulva prolifera green tide and increasingly complex marine operating environment, crew safety supervision faces severe challenges. For salvage operations in which there are violations such as not wearing helmets and life jackets, the traditional manual supervision means are lagging and subjective, making it difficult to realize efficient and accurate identification. Therefore, this paper improves the YOLOv8 model based on the 2024 ulva prolifera salvage surveillance images to construct a violation operation identification and detection model. For the problems of difficult detection of small targets and large scale changes, a deformable convolutional space pyramid pooling module and a UES fusion attention mechanism module are proposed to improve the detection accuracy, named YOLOv8s-DU. To further improve the operational efficiency, a C2f-Faster module and a lightweight structure of GSConv are adopted to reduce the computational overhead under the premise of guaranteeing the precision and enhance the real-time response capability, and the final model is named YOLOv8s-DUFG. By experiment, it proves the effectiveness of the improved algorithm in this paper in the context of ulva prolifera salvage illegal operation identification, which provides technical support for the marine regulatory departments in the supervision of maritime operation behavior and emergency warning, and also provides a reference scheme for the improvement of the illegal operation identification model.