为了保障航运安全，提出基于机器视觉技术的船舶吃水深度动态检测。将预处理后的船舶SAR图像作为I-VGGNet网络的输入，通过I-VGGNet网络的卷积层结构提取船舶SAR图像不同层次的特征。在此基础上，使用FCOS网络对船舶SAR图像特征进行尺度划分，再引入IoU损失函数和RCIoU损失函数获得预测框和真实框最小的中心距离，以此校正船舶的水尺字符印，确定吃水线，实现船舶吃水深度检测。实验结果表明，该方法能够准确校正船舶的水尺字符印，并精准识别不同尺度下的船舶目标。且总体AP值为95.8%，相对较高，可以有效检测船舶吃水深度，保证船舶的安全。
In order to ensure shipping safety, a dynamic detection of ship draft depth based on machine vision technology is proposed. Using the preprocessed ship SAR image as input to the I-VGGNet network, the convolutional layer structure of the I-VGGNet network is used to extract features at different levels of the ship SAR image. On this basis, the FCOS network is used to scale the ship SAR image features, and then the IoU loss function and RCIoU loss function are introduced to obtain the minimum center distance between the predicted box and the true box, in order to correct the ship's draft character print, determine the waterline, and achieve ship draft depth detection. The experimental results show that this method can accurately correct the water gauge character marks of ships and accurately identify ship targets at different scales. And the overall AP value is 95.8%, relatively high, which can effectively detect the draft depth of ships and ensure their safety.
2024,46(14): 158-161 收稿日期：2024-01-19
DOI：10.3404/j.issn.1672-7649.2024.14.026
分类号：TN959
基金项目：开封大学青年创新人才培育基金资助项目（KDQN-2020-GK008）；河南省博士后科研启动基金资助项目（HN2022105）；河南省科技攻关项目（232103810059）
作者简介：冯维娜(1988-),女,硕士,讲师,研究方向为人工智能与计算机应用
参考文献：
[1] 贾程澄, 王军, 邱峰. 多特征提取和多核SVM的舰船目标识别方法[J]. 电光与控制, 2021, 28(11): 106-111.
[2] 陈轩, 周家海. 船舶吃水深度数据检测关键技术及应用[J]. 舰船科学技术, 2022, 44(8): 182-185.
CHEN Xuan, ZHOU Jiahai. Key technology and application of ship draft depth data detection[J]. Ship Science and Technology, 2022, 44(8): 182-185.
[3] 佟世琪, 赵鹏, 王月兵, 等. 基于超声相控技术的船舶吃水测量方法[J]. 声学技术, 2023, 42(2): 192-198.
[4] 饶本顺, 郑云峰, 周校彬, 等. 散装谷物船自动配载方案分析[J]. 船海工程, 2022, 15(5): 113-116+125.
[5] CHAKRABORTY S , AMRITA, CHOUDHURY T , et al. Multi-view deep cnn for automated target recognition and classification of synthetic aperture radar image[J]. Journal of Advances in Information Technology, 2022, 13(5): 413–422.
[6] 马剑英, 朱迪. 基于星载宽刈幅散射计的海面舰船目标检测方法与识别[J]. 遥感技术与应用, 2021, 36(3): 663-672.
[7] 陈瑾龙, 徐哲壮, 黄平, 等. 基于机器视觉的四足机器人目标识别与跟随系统设计[J]. 实验技术与管理, 2022, 39(10): 135-139+167.
[8] 雷禹, 冷祥光, 周晓艳, 等. 基于改进ResNet网络的复数SAR图像舰船目标识别方法[J]. 系统工程与电子技术, 2022, 44(12): 3652-3660.
