Multiple Attention Mechanism Enhanced YOLOX for Remote Sensing Object Detection.

Area of Science:

• Computer Science
• Remote Sensing
• Artificial Intelligence

Background:

• Object detection in remote sensing is crucial for environmental monitoring, urban planning, and defense.
• Current algorithms struggle with detecting small objects in complex backgrounds, limiting their robustness.

Purpose of the Study:

• To propose a novel algorithm, Multiple Attention Mechanism Enhanced YOLOX (MAME-YOLOX), for robust tiny object detection in remote sensing.
• To enhance the focus on salient information and improve the perception of local geometric features.

Main Methods:

• Integrated the CBAM attention mechanism into the YOLOX backbone to highlight saliency.
• Incorporated the Swin Transformer into the YOLOX neck for high-level semantic and local geometric feature perception.
• Utilized CIoU loss for bounding box regression to prevent degeneration.

Main Results:

• The MAME-YOLOX algorithm demonstrated superior performance on AIBD, HRRSD, and DIOR datasets.
• Quantitative and qualitative experimental results confirmed the algorithm's effectiveness.
• The proposed method shows significant improvements in detecting tiny objects against complex backgrounds.

Conclusions:

• The MAME-YOLOX algorithm effectively addresses the limitations of existing object detection methods for tiny objects in remote sensing.
• The integration of attention mechanisms and Swin Transformer significantly enhances detection capabilities.
• This research contributes a more robust and accurate solution for remote sensing object detection applications.