MSA-YOLO: A Remote Sensing Object Detection Model Based on Multi-Scale Strip Attention.

Area of Science:

• Computer Vision
• Remote Sensing Technology
• Artificial Intelligence

Background:

• Object detection in remote sensing is crucial for resource and environmental monitoring.
• Challenges include complex backgrounds and significant variations in object sizes.
• Existing models struggle with these complexities, necessitating improved detection methods.

Purpose of the Study:

• To propose an efficient remote sensing image object detection model, MSA-YOLO.
• To enhance detection performance by addressing background noise and multi-scale object variations.
• To improve model efficiency and accuracy for practical applications.

Main Methods:

• Developed a Multi-Scale Strip Convolution Attention Mechanism (MSCAM) to reduce background noise and fuse multi-scale features.
• Integrated the lightweight GSConv module and an improved feature fusion layer for a more efficient and accurate model.
• Introduced a Wise-Focal CIoU loss function to balance sample contributions and improve regression.

Main Results:

• The proposed MSA-YOLO model demonstrated significantly superior performance compared to existing methods.
• Experiments were conducted on public remote sensing datasets, including DIOR and HRRSD.
• The model effectively handles complex backgrounds and objects of varying sizes.

Conclusions:

• MSA-YOLO offers an efficient and accurate solution for remote sensing image object detection.
• The novel attention mechanism, lightweight design, and improved loss function contribute to enhanced performance.
• This model shows great potential for applications in resource and environmental management.