基于曼哈顿距离加权协同表示分类的车辆识别

Vehicle Recognition based on Manhattan Distance Weighted Collaborative Representation based Classification

摘要:

加权稀疏表示分类（WSRC）在声频传感器网络下的车辆识别中取得了不错的效果。但是稀疏表示分类（SRC）中实际上起较大作用的是字典中所有类的协同表示，因此协同表示分类（CRC）被提出用来提升算法效率，CRC框架还改进了残差计算方式来提高识别精度。在WSRC中发现保局性对提升识别率起到很好的作用，因此在CRC中引入加权编码，提出了声频传感器网络下基于加权协同表示分类（WCRC）的车辆识别方法，取得了明显的速度（相比WSRC、SRC）以及不错的精度（对比WSRC、CRC、SRC）提升。同时针对欧氏距离对样本相似性判断的不足，将曼哈顿距离引入加权编码，进一步地提出了基于曼哈顿距离加权协同表示分类（Manhattan-WCRC）的车辆识别方法，取得了最高的识别率，而运算速度与WCRC接近。

Abstract:

Weighted Sparse Representation based Classification (WSRC) has achieved good results in vehicle recognition in acoustic sensor networks. However, the collaborative representation of all classes in the dictionary actually plays an important role in the Sparse Representation based Classification (SRC).Collaborative Representation based Classification (CRC) is proposed to improve the efficiency of the algorithm. The CRC framework also improves the residual calculation method to improve the recognition accuracy.It is found in WSRC that data locality plays a very good role in improving the recognition rate. Therefore, weighted coding is introduced into CRC, and a vehicle recognition method based on Weighted Collaborative Representation based Classification (WCRC) in acoustic sensor networks is proposed, which achieves obvious speed (compared with WSRC, SRC) and good accuracy (compared with WSRC, CRC, SRC) improvement.At the same time, in view of the shortcomings of Euclidean distance in judging sample similarity, the Manhattan distance is introduced into weighted coding, and a vehicle recognition method based on Manhattan distance Weighted Collaborative Representation based Classification(Manhattan-WCRC) is proposed. Manhattan-WCRC achieves the best recognition rate with almost the same speed as WCRC.