提出了一种基于光谱分析理论和张量代数理论的高分辨率遥感影像目标探测器。首先在向量空间中对目标光谱进行特征匹配;然后对光谱特征与目标相似的像素进行空间-光谱特征一体化张量描述,进而在张量特征空间中对目标和背景进行分类。实验表明,张量学习机能够有效地对高分辨率影像进行目标识别。加入光谱特征匹配后的目标探测器能够极大地降低目标探测时间,同时进一步提高目标探测精度。
We proposed a hybrid detector based on spectral matching and tensor analysis, which is designed for hyperspectral and high resolution remote sensing images. Firstly, a spectral matching is performed in vector space using adaptive coherence/cosine estimator (ACE). Then, the result pixels of whose spectral are similar with targets spectral are further processed by support tensor machine (STM) to detect the real targets from the background pixels. The experimental results on CRI dataset demonstrate that the proposed approach could obviously reduce the processing time and improve the targets detection precision.