MSPPF-NETS: A DEEP LEARNING ARCHITECTURE FOR REMOTE SENSING IMAGE CLASSIFICATION

Nowadays, deep learning has got a major success in computer vision, especially in image recognition. In this paper, a new architecture based on DenseNets which is referred to as Multi-Scale Input Spatial Pyramid Pooling Fusion Networks (MSPPF-nets) is proposed for the work of classification of local climate zones (LCZs). Multi-scale remote sensing images can be inputs of the networks by the benefit of Spatial Pyramid Pooling (SPP) layer, multi-scale features from different channels were extracted and fused by our multi-branch-input framework. The final classification results have illustrated the feasibility of this presented classification method.