Remote sensing image scene classification based on multiscale features fusion

Abstract

In order to solve the difficulties of multi-scale feature extraction and weak representation in remote sensing image scene classification, a classification method based on multi-scale feature fusion (MFF) was proposed. The convolutional representation and fully connected features generated by the feature fusion of the MFF are used as high-level features to generate discriminative scene representations, which are then input into the softmax classifier to obtain the semantic labels of scenes. The existing convolutional neural network-based methods and MFF methods are tested on three widelyused datasets. The results show that the MFF method has higher overall accuracy than the existing convolutional neural network-based methods and can better meet the current demand for remote sensing image scene classification.