A Lightweight Patch-Level Change Detection Network Via Exploring The Potential of Pruning and Multi-Scale Pooling
Lihui Xue, Xueqian Wang, Zhihao Wang, Linping Zhang, Gang Li · 2024
Existing satellite remote sensing change detection (CD) methods often crop large-scale bi-temporal image pairs into small patch pairs and use pixel-level CD methods to fairly process all the patch pairs. However, due to the sparsity of changed areas, existing pixel-level CD methods suffer from a waste of computational cost and memory resources on many unchanged areas, which hinders the deployment of the CD model on on-board platforms with extremely limited resources. To address this issue, we propose a lightweight patch-level CD network (LPCDNet) to rapidly remove the unchanged patch pairs in large-scale bi-temporal optical image pairs, which is helpful to accelerate the subsequent pixel-level processing and reduce its memory costs. In LPCDNet, based on the multi-scale max-pooling structure, the multilayer feature compression (MLFC) module is designed to compress and fuse the multi-level feature information from backbone network. Moreover, a sensitivity-guided network pruning method is proposed to remove unimportant channels and construct a lightweight backbone network based on ResNet18. Experiments on two datasets demonstrate the effectiveness and efficiency of our proposed method compared with existing methods.