基于深度学习的红外烟幕区域分割技术  

A Method of Smoke Area Segmentation for Infrared Images Based on Deep Learning

作　　者：齐航 袁健全 李磊 任君 梁杰 QI Hang;YUAN Jianquan;LI Lei;REN Jun;LIANG Jie(Science and Technology on Complex System Control and Intelligent Agent Cooperation Laboratory, Beijing 100074, China;Institute of Mechanical and Electrical Engineering, Beijing 100074, China)

机构地区：[1]复杂系统控制与智能协同技术重点实验室,北京100074 [2]北京机电工程研究所,北京100074

出　　处：《控制与信息技术》2019年第4期18-22,57,共6页CONTROL AND INFORMATION TECHNOLOGY

基　　金：国家自然科学基金(61803356);国防基础科研计划(JCKY2017204B064)

摘　　要：红外成像制导系统容易受到烟幕干扰,分割烟幕区域是提高红外成像自动目标识别性能的有效技术途径。文章提出一种基于深度学习的红外烟幕干扰分割技术,针对实拍红外烟幕图像样本获取成本高而难以满足算法训练需求的问题,提出基于粒子系统的红外烟幕仿真技术,并针对性地借助图像增广技术模拟不同成像条件,借助Deeplab v3+算法实现烟幕区域分割。文章在零实拍红外图像条件下训练网络,在实拍图像集上测试结果与真实烟幕区域的平均交并比达到79%,实现了红外图像烟幕区域分割。Smoke area segmentation is an efficient way to improve the performance of the infrared imaging guidance system, even though smoke is one of the main disturbances for automatic target recognition. In this paper, we proposed a smoke area segmentation method for infrared images based on deep learning. In order to reduce the cost of data acquisition, the deep learning network for cloud segmentation is trained on our simulated image data set up by particle system and data augmentation, which substantially decreases the demand for real smoke images. Then, the Deeplab v3+ algorithm was used for smoke segmentation. When testing on the real image data set, a satisfactory segmentation result with 79 percent has been reached which can meet the usage requirements, illustrating that our method is effective and efficient.

关 键 词：深度学习 图像语义分割 红外图像 粒子系统 图像增广 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]