基于深度学习的水下移动目标快速识别方法研究  

Research on fast recognition method of underwater moving target based on deep learning

作　　者：盛德奎 SHENG Dekui(Unit 92213,People's Liberation Army,Zhanjiang Guangdong 524064,China)

机构地区：[1]中国人民解放军92213部队,广东湛江524064

出　　处：《自动化与仪器仪表》2021年第12期8-11,共4页Automation & Instrumentation

基　　金：国家自然科学基金:51972568。

摘　　要：由于传统的水下移动目标快速识别方法不能精准且快速地识别水下移动目标,提出一种基于深度学习的水下移动目标快速识别方法。采用深度学习中的神经网络进行知识模型训练,在此基础上,根据点扩散函数构建水下光学环境模型,并通过标定图学习获取光学参数获取,同时为降低变化畸形,使用优化后Hu氏不变矩在去除干扰基础上获取图像特征,并凭借反解径向畸变模型,输出映射像素坐标间理想的对应关系,最后利用像素分辨率完成水下移动目标识别。实验结果表明,基于深度学习的水下移动目标快速识别方法能够精准地识别水下移动目标,且提高了识别效率。Because the traditional rapid recognition method of underwater moving target cannot accurately and quickly recognize underwater moving target,a rapid recognition method of underwater moving target based on deep learning is proposed.The neural network in deep learning is used for knowledge model training.On this basis,the underwater optical environment model is constructed according to the point spread function,and the optical parameters are obtained through calibration map learning.The torque conversion obtains the image features on the basis of removing the interference,and uses the inverse solution of the radial distortion model to output the ideal correspondence between the mapped pixel coordinates,and finally uses the pixel resolution to complete the underwater moving target recognition.The experimental results show that the fast recognition method of underwater moving targets based on deep learning can accurately recognize underwater moving targets and improve the recognition efficiency.

关 键 词：深度学习 目标识别 Hu氏不变矩 点扩散函数模型 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置] TM214[自动化与计算机技术—控制科学与工程]