Combining Transformers and CNNs for Efficient Object Detection in High-Resolution Satellite Imagery

2025-07-15

AI Summary

To address weak feature representation and inefficient cross-scale fusion in high-resolution satellite image object detection, this paper proposes GLOD: a novel architecture that replaces the conventional CNN backbone with Swin Transformer to enhance long-range dependency modeling; introduces an UpConvMixer upsampling module and a multi-scale Fusion Block for efficient feature reconstruction and fusion; innovatively adopts an asymmetric cross-layer fusion strategy incorporating CBAM attention mechanisms; and employs a multi-path detection head to strengthen multi-scale object representation. GLOD significantly improves the exploitation of spatial priors inherent in satellite imagery. Evaluated on the xView dataset, it achieves 32.95% mAP—surpassing the previous state-of-the-art by 11.46%—while attaining a superior trade-off between detection accuracy and computational efficiency.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

We present GLOD, a transformer-first architecture for object detection in high-resolution satellite imagery. GLOD replaces CNN backbones with a Swin Transformer for end-to-end feature extraction, combined with novel UpConvMixer blocks for robust upsampling and Fusion Blocks for multi-scale feature integration. Our approach achieves 32.95% on xView, outperforming SOTA methods by 11.46%. Key innovations include asymmetric fusion with CBAM attention and a multi-path head design capturing objects across scales. The architecture is optimized for satellite imagery challenges, leveraging spatial priors while maintaining computational efficiency.

Problem

• Improving object detection in high-resolution satellite imagery
• Combining Transformers and CNNs for efficient feature extraction
• Addressing multi-scale object detection with novel fusion blocks

Innovation

• Swin Transformer replaces CNN backbones
• UpConvMixer blocks for robust upsampling
• Asymmetric fusion with CBAM attention