Heterogeneous Image Template Matching Based on Region Proposal Network
Sha Yue, Wang Shengzhe, Yuyong Cui, Wei Guan, Xinyi Gao · Journal of Physics Conference Series · 2021
Abstract Heterogeneous template matching is important to disaster relief. The method of traditional heterogeneous template matching in real-time performance is so weak and usually has a lower accuracy than deep learning method. A heterogeneous template matching method based on region proposal network is proposed in this paper. Specifically, deep features between the optical images and SAR images are extracted by using residual network, and MRPN, a suitable region proposal network for matching work, is proposed to complete precise positioning of the template region. Finally, the channel fusion is performed on the matching results to get a better result and confidence. It is shown that the real-time performance and the accuracy performance have been greatly improved, which is reaching 97%.