基于改进YOLOv3目标检测算法的船舶运载货物自动识别研究

打开文本图片集

摘要：船舶货物自动识别高精度数据获取难，影响检测性能。该文利用弱监督至全监督框架，结合改进算法构建组合框架，平均识别精度达32.0%，定位精度达73.8%，高于对比方法。该框架在弱监督环境下表现优异，适用于船舶货物自动识别。

关键词：YOLOv3；弱监督；船舶运载；候选区域

doi：10.3969/J.ISSN.1672-7274.2024.09.001

中图分类号：TP 391.41                 文献标志码：A            文章编码：1672-7274（2024）09-000-03

Research on Automatic Identification of Ship Cargo Based on Improved YOLOv3 Object Detection Algorithm

HOU Guojiao1， SUN Rong1， XIAO Shengkui1， LI Wen1， ZHANG Dong2

（1. The Navigation Authority of Yangtze Gorges， Yichang 443002， China;

2. Hunan Tianxiakuan Information Technology Co.， Ltd.， Changsha 410000， China）

Abstract： The difficulty in obtaining high-precision data for automatic identification of ship cargo affects the detection performance. This study utilizes a weak supervision to full supervision framework combined with improved algorithms to construct a combined framework. The average recognition accuracy reaches 32.0%， and the positioning accuracy reaches 73.8%， which is higher than the comparison methods. This framework performs excellently in a weak supervision environment and is suitable for automatic identification of ship cargo.

Keywords： YOLOv3; weak supervision; ship transportation; candidate region

0   引言

在船舶货物自动识别领域，视觉图像的检测识别扮演着核心角色，而人工智能算法的兴起为此提供了新的思路[1]。（剩余3662字）