基于非完备字典的舰船特征提取和识别

Feature Extraction and Recognition of Ships by an Uncompleted Dictionary

提出了一种基于非完备特征字典的舰船特征提取和识别算法。借鉴稀疏表示理论的思想,根据红外图像中舰船的外形特征数据集构造特征字典,将目标信号进行分解,根据匹配字典中每个特征基原子得到特征响应,从而获得目标的特征表示,最后采用SVM一对一投票方法进行目标识别,得到最终的目标识别结果。仿真实验表明,与简单的标量区域描述方法和矩特征方法相比,本文方法得到的特征不仅具备更快的提取速度,而且可以更好地区分目标,提高目标识别的正确率。

This paper presents a novel approach of extracting ships feature by an uncompleted dictionary. First, we refer to the thought of sparse representation, and the uncompleted dictionary is constructed in terms of various ships shape in infrared images. Furthermore, using orthogonal matching pursuit algorithm to decompose the ship target signal, we obtain a ship response that describes the target feature. Finally, the classification result is decided by voting strategy. The results of experimentation indicate that the proposed approach has better performance than moments feature, compactness, rectangularity and so on.