多尺度超完备字典的飞机目标检测算法  

Aircraft Targets Detecting Algorithm Based on Overcomplete Dictionary in Multi-scale

作　　者：陈滨[1] 赵建军[1] 王毅[1] 姚刚[1] 

机构地区：[1]海军航空工程学院兵器科学与技术系,烟台264001

出　　处：《科学技术与工程》2017年第17期105-110,共6页Science Technology and Engineering

摘　　要：天空背景的复杂性与飞机目标的多样性,对基于传统目标检测算法的飞机目标检测带来了巨大的挑战。按图像的稀疏表示理论,提出了多尺度超完备字典的飞机目标检测算法。算法综合了不同尺度下超完备字典各自的优点:利用低分辨率图像块学习小尺度字典,构造小尺度分类器,在低分辨率测试图像中完成前景粗检测;利用高分辨率图像块学习大尺度字典,构造大尺度分类器,在高分辨率测试图像中完成前景精检测;最后通过飞机图像块学习飞机目标字典,构造飞机目标分类器,完成前景目标分类。实验结果表明,算法能有效完成天空背景下的飞机目标检测任务,并在耗时、检测命中率、误检率等方面均有良好表现。Aircraft targets detection based on traditional object detecting algorithm encounter problems caused by the complexity of sky background and diversity of aircrafts. A detecting algorithm of aircraft targets based on overcomplete dictionary with multi-scale in the domain of sparse representation was proposed. The proposed algo-rithm synthesizes the advantages of overcomplete dictionary in different scales. The dictionary in small scale is trained by image blocks with low resolution. The classifier in small scale is constructed to complete the cursory de-tection of foreground. In the similar way, the dictionary in large scale is trained by image blocks with high resolu-tion. Then, the classifier in large scale is constructed to complete the refined detection of foreground. The aircraft targets classifier completes the classification of foreground objects. As the experiments show, the proposed algorithm completed the task of aircraft targets detection efficiently, and had good performance in time consumption, as well as in hit rate and false drop rate.

关 键 词：天空背景 飞机目标检测 图像稀疏表示 多尺度 超完备字典 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]