ELI-Net: Ensemble learning interpretable networks for the classification of remote sensing images

Fine-grained target classification in optical remote sensing images is becoming one of the hotspots in the field of remote sensing image classification. Although many deep learning models have demonstrated good performance in fine-grained classification tasks for remote sensing images, the complexity and poor interpretability of the models have seriously affected the application of related techniques, and many existing interpretability analyses come at the expense of performance. In light of this, we propose an ensemble learning interpretable network (ELI-Net) for fine-grained classification in optical remote sensing images. The specific process is to provide interpretable analysis by transforming a common classification network structure and to improve the model accuracy by ensemble learning. We also carry out experiments on the FGSCR-42, a dataset of the fine-grained ship classification in remote sensing images, and the experimental results reveal that the transformed network model can achieve the goal of providing visual interpretation while ensuring higher performance.