Abstract: Target identification from hyperspectral images is one of its most important applications, which wants to find out special targets from background. Researchers have taken full use of spectral and spatial information to increase the accuracy of target identification. However, the diversity and complexity of targets make it hard to obtain obvious improvement with current technologies. Now the development of earth observation technology makes it possible to acquire multi-temporal hyperspectral images covering the same area more frequently. So temporal information mining can be a feasible solution to the bottleneck.
Temporal information mining is to find out temporal information from multi-temporal hyperspectral data to extract temporal features of targets, and thus improve identification ability. The targets in study scene mostly have their special temporal patterns during the observation, for example, tanks and planes will appear and move from one time to another time. Temporal information mining can help us automatically locate the special target with special temporal pattern, and finally identify what we want combined with spectral and spatial information. Several advanced theories available in temporal information mining, such as deep learning and semantic analysis, will be presented. The real applications and future development for temporal information mining will also be addressed.