SWIN-Fovea: A Remote Sensing Object Detection Model Based on Swin Transformer and FoveaBox
Yunxiao Lin, Jianguo Hu, Yanyu Ding, Miaowang Zeng, Chanzi Liu · 2023
Recently, convolutional neural networks (CNNs) have achieved excellent results in object detection of remote sensing images. However, pure CNN-based object detection models have difficulty in learning explicit long-range relational modeling due to the inherent locality of convolutional operations. To solve these problems, we proposed a hybrid architecture for object detection by combining transformers and CNNs. In which the backbone network employs a Swin Transformer-based FPN, while the detection head comprises a CNN-based FoveaBox classifier and regressor. We named the proposed model Swin-Fovea and conducted extensive experiments on the DIOR dataset for remote sensing object detection. The experimental results suggested the superiority of our approach over other advanced methods.