基于多尺度核索引字典的飞机目标检测优化仿真  

OPTIMIZATION AND SIMULATION OF AIRCRAFT TARGET DETECTION BASED ON MULTI-SCALE KERNEL INDEX DICTIONARY

作　　者：陈滨[1] 赵建军[1] 杨利斌[1] 王毅[1] 

机构地区：[1]海军航空工程学院兵器科学与技术系,山东烟台264001

出　　处：《计算机应用与软件》2017年第11期197-203,222,共8页Computer Applications and Software

摘　　要：为进一步提高基于图像稀疏表示的飞机目标检测算法的时间性能与精确度,提出了基于多尺度核索引字典的飞机目标检测算法,分别从超完备字典结构、目标检测分类器结构两方面优化算法。首先引入基于高斯径向核函数的硬C聚类方法,构造核索引字典,在提升稀疏求解算法时间性能的同时,提高了索引字典原子聚类的准确度。接着基于核索引字典,构建多尺度分类器,进一步提高了算法的效率和精度。实验表明,在合理选择聚类数后,采用核索引字典有效降低了稀疏求解算法的时间开销,原子的聚类准确度有所提高;相对基于单尺度字典的飞机目标检测算法,基于多尺度核索引字典的算法在时间开销上平均降低至24.7%,在精度方面,误检率平均降低了20.3%,命中率平均提高了3.4%,满足实时应用要求。In order to improve the time performance and accuracy of aircraft target detection algorithm, we propose an aircraft target detection algorithm based on multi-scale kernel index dictionary. The algorithm is optimized from the construction of dictionary and classifies of object detection. First, the RBF kernel was introduced into the HCM algorithm to construct the indexed dictionary. Time performance was improved as well as the accuracy of clustering. Then, the multi-scale classifier was constructed based on the kernel index dictionary to further improve the efficiency and accuracy of the algorithm. As experiments show, after choosing a reasonable number of clusters, kernel-based indexed dictionary has decreased the time consumption of sparse solution. The accuracy of clustering has increased at the same time. Compared with the single scale dictionary, the algorithm based on the multi-scale kernel index dictionary reduces the time cost to 24.7%. In the respect of accuracy, the false detection rate decreased by an average of 20.3%, and the average hit rate increased by 3.4%. In conclusion, the proposed algorithm can satisfy the requirement of real-time application.

关 键 词：飞机目标检测 核聚类 索引字典 多尺度 稀疏表示 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]