YOLO-HVS: Infrared Small Target Detection Inspired by the Human Visual System

Area of Science:

• Computer Vision
• Artificial Intelligence
• Infrared Imaging

Background:

• Infrared small target detection faces challenges from background interference.
• Limited multi-scale feature extraction hinders performance in complex scenarios.

Purpose of the Study:

• To develop an efficient and robust algorithm for infrared small target detection.
• To improve detection accuracy under severe occlusion and low signal-to-noise ratio (SNR) conditions.

Main Methods:

• Proposes YOLO-HVS, an algorithm based on YOLOv8, incorporating a multi-scale spatially enhanced attention module (MultiSEAM).
• Introduces the C2f_DWR module with a regional-semantic dual residual structure for efficient multi-scale contextual information capture.
• Constructs the DroneRoadVehicles dataset with 1028 infrared images featuring complex occlusions and multi-scale targets.

Main Results:

• YOLO-HVS achieved mAP50 of 83.4% on DroneVehicle and 97.8% on the DroneRoadVehicles dataset.
• Demonstrated improved performance over baseline YOLOv8 with minimal increases in model parameters and GFLOPs.
• Exhibited enhanced robustness in detecting targets under severe occlusion and low SNR.

Conclusions:

• The proposed YOLO-HVS algorithm effectively addresses background interference and enhances multi-scale feature extraction for infrared small target detection.
• The algorithm offers improved robustness and efficiency for real-time applications.
• The developed DroneRoadVehicles dataset provides valuable data for advancing research in this field.