In response to the current ship detection methods that rely on conventional object detec-tion techniques and require non-maximum suppression(NMS)post-processing,an enhanced end-to-end object detection method is proposed,which generates high-quality proposal boxes and fea-tures using an adaptive proposal network and utilizes a prediction update network to improve the prediction results.The method has been experimentally verified to improve detection accuracy without changing the end-to-end nature,effectively handling complex environments,small vessels,and fine-grained classification,providing a basis for related research.
维普
万方数据