关键词: 目标检测, 多模态航天遥感图像, 混合专家组, 自蒸馏, 傅里叶变换, 扩散模型

Abstract: With the number of remote sensing satellites deployed in orbit in China increasing year by year, the quantity of aerospace remote sensing images, represented by Synthetic Aperture Radar (SAR) and optical (RGB) images, is rapidly growing, and the demand for tasks such as object detection from these massive datasets is also rapidly rising. However, due to objective factors such as imaging mechanisms and resolution differences, there are significant modality feature differences between images from different satellites. This difference is particularly pronounced between SAR and RGB remote sensing images, making it difficult for a single model to learn features from different types of remote sensing images. Consequently, each satellite requires a dedicated model for detection tasks, which has become a major obstacle to collaborative recognition and relay detection applications in satellite remote sensing. To address this issue, this paper innovatively proposes a self-distilled multimodal detection model based on a Mixture of Experts (MoE) network. First, a modality-aware MoE structure is constructed, using a small number of high-quality experts as teachers to guide the other experts, while simultaneously incorporating modality-invariant constraints to further reduce cross-modality feature shifts. Second, a Fourier-enhanced diffusion detection head is constructed and combined with frequency-domain feature enhancement to improve the ability to capture detailed information about detection targets. To evaluate the model's performance, aerospace images were selected and cropped from the public datasets FAIR1M and SARDet_100K, resulting in a dataset of 68,983 aerospace remote sensing images for object detection under different backgrounds and imaging mechanisms. Experimental results show that, compared with existing single-modality detection methods, the proposed model performs better in detection tasks across both modalities, with a significant improvement in mean average precision (mAP). This fully demonstrates that the proposed model has good application value in multimodal aerospace remote sensing image object detection and is well-suited for various types of satellite remote sensing images.

Key words: Object detection, multimodal aerospace remote sensing images, Mixture of Experts, self-distillation, Fourier transform, diffusion model