An Empirical Study of Remote Sensing Pretraining

Deep learning has largely reshaped remote sensing (RS) research for aerial image understanding and made a great success. Nevertheless, most of the existing deep models are initialized with the ImageNet pretrained weights since natural images inevitably present a large domain gap relative to aerial images, probably limiting the fine-tuning performance on downstream aerial scene tasks. This issue motivates us to conduct an empirical study of RS pretraining (RSP) on aerial images. To this end, we train different networks from scratch with the help of the largest RS scene recognition dataset up to now—MillionAID—to obtain a series of RS pretrained backbones, including both convolutional neural networks (CNNs) and vision transformers, such as Swin and ViTAE, which have shown promising performance on computer vision tasks. Then, we investigate the impact of RSP on representative downstream tasks, including scene recognition, semantic segmentation, object detection, and change detection using these CNN and vision transformer backbones. Empirical study shows that RSP can help deliver distinctive performances in scene recognition tasks and in perceiving RS-related semantics, such as “Bridge” and “Airplane.” We also find that, although RSP mitigates the data discrepancies of traditional ImageNet pretraining on RS images, it may still suffer from task discrepancies, where downstream tasks require different representations from scene recognition tasks. These findings call for further research efforts on both large-scale pretraining datasets and effective pretraining methods. The codes and pretrained models will be released at https://github.com/ViTAE-Transformer/ViTAE-Transformer-Remote-Sensing.

I. Introduction

With the development of geoinformatics technology, the Earth observation fields have witnessed significant progress, where various remote sensing (RS) sensors and devices have been widely used. Among them, with the advantages of real-time, abundant amount, and easy access, the aerial image has become one of the most important data sources in Earth vision to serve the requirements of a series of practical tasks, such as precision agriculture [1], [2] and environmental monitoring [3]. In these applications, aerial scene recognition is a fundamental and active research topic over the past years. However, because of the own characteristics of aerial images, it is still challenging to efficiently understand the aerial scene.