提出了一种基于张量学习机的遥感影像目标探测方法。该方法基于张量数据模型和张量代数运算, 针对遥感影像数据多维或高维的特点, 将基于向量的监督法学习机扩展为基于张量的监督法学习机, 然后利用凸函数最优化理论和交互投影迭代法求得张量学习机的最优解。最后分别以高光谱遥感影像和高分辨率遥感影像为例, 使用张量学习机进行目标探测。实验表明, 与支持向量机等方法相比, 本文的方法在保持较高探测成功率的同时更好的抑制了虚警。

This paper proposes a new way to detect the targets in remote sensing images based on the tensor learning machine (TLM). This method is based on tensor and tensor algebra. To utilize the multidimensional data of the remote sensing image, the vector-based learning machine is generalized to the tensor-based learning machine which accepts tensors as input, then the convex optimization theory and the alternating projection procedure are used to get the solution of the TLM. TLM is tested to target detection using the hyperspectral remote sensing data and high resolution remote sensing data. The experiments demonstrate the effectiveness of the proposed method, by comparing TLM with support vector machine, the tensor learning machine can keep a high probability of successful detection and reduce the false alarm.