MFFD: Multilayer Feature Fusion and Decision Network for Remote Sensing Image Classification

In recent years, with the rapid development of computer vision technology, remote sensing image classification has gained increasing attention. Nevertheless, due to the high resolution and rich scale information of remote sensing images, existing lightweight deep learning models often struggle to fully capture their detailed and multi-scale features, while large-scale CNNs and Transformers face challenges with massive parameter sizes and high computational costs. To address the above issues, we propose a Multilayer Feature Fusion and Decision (MFFD) based on EfficientNet B0. This method starts by extracting middle layer and deep layer features to effectively capture the spatial details and high-level semantics of remote sensing images. The Middle Layer Feature Enhancement (MLFE) and Deep Layer Feature Enhancement (DLFE) modules are designed to process these features separately, enhancing both detail preservation and semantic expressiveness. Finally, features from different layers are fused through concatenation and subjected to decision-level classification using multiple sub-classifiers, thereby enhancing classification accuracy and robustness. Experimental results show that MFFD achieves high classification accuracy across multiple remote sensing image benchmark datasets as well as demonstrating advantages in computational efficiency and parameter optimization. Habitually, the MFFD's source code can be made publicly accessible on my profile page at https://github.com/li-zi-qi/MFFD .