基于端到端的船舶检测算法  

An end-to-end ship detection algorithm

作　　者：毛硕 徐家辉 冯晓慧 王保兴 MAO Shuo;XU Jiahui;FENG Xiaohui;WANG Baoxing(The Eighth Research Academy of CSSC,Nanjing 211153;Unit 91177 of the PLA,Qingdao 266000,China)

机构地区：[1]中国船舶集团有限公司第八研究院,南京211153 [2]91197部队,山东青岛266000

出　　处：《雷达与对抗》2024年第4期30-33,共4页Radar & ECM

摘　　要：针对现有船舶检测方法依赖传统的目标检测技术,须进行非极大值抑制后处理的问题,提出一种增强的端到端的目标检测方法,用自适应提议网络生成高质量的提议框和提议特征,利用预测更新网络对预测结果进行改进。通过实验验证了该方法能够在不改变端到端的前提下提高检测精度,有效处理复杂环境、小型船只和细粒度分类,可为相关研究提供依据。In response to the current ship detection methods that rely on conventional object detection 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 features 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.

关 键 词：端到端 细粒度 目标检测 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]