Preferred Feature Representation for Remote Sensing Scene Classification

Meng Yang, Yaxiong Chen, Xiongbo Lu, Shengwu Xiong

Abstract: For remote sensing image analysis field, remote sensing scene classification is the main task to label the semantic content of remote sensing images. Remote sensing scene images have highly complex geometric structures. Convolutional neural networks can only capture short-range dependencies and cannot adapt to the unique spatial patterns of remote sensing images, leading to insufficient information extraction for more discriminative features. Remote sensing images also face difficulties such as large intra-class variability and high inter-class similarity. To address these challenges, we propose a new method for efficiently learning feature representations using two modules: a Feature Preference Module (FPM) for single image data and a Batch Cross-sample Attention Module (BCAM) for multiple image data. This approach aims to enhance the discriminative power of features. We conducted experiments on the ERA, AID, and UC Merced datasets, achieving state-of-the-art performance and demonstrating the effectiveness of our proposed method.