An Underwater Spam Detection Algorithm based on Enhanced YOLOv8

The present paper proposes an enhanced YOLOv8-based method for detecting underwater debris. The lightweight model YOLOv8n is selected as the baseline and optimized and improved in this way, our model can address the issues of inadequate detection, slow processing speed, and excessive model size in the previous approach. The paper enhances the model’s ability to focus on key features by incorporating the high-efficiency attention module, thereby enhancing model performance in complex underwater environments. Following the investigation into Ghost convolution and subsequent comparative tests, it has been resolved to substitute the standard convolution with Ghost convolution in the model’s neck region. This modification aims to enhance the model’s inference velocity and render the model more compact. Ultimately, Comparison and ablation studies on public data sets demonstrate the effectiveness of the proposed method and advancement of the method.