YOLO-B:An infrared target detection algorithm based on bi-fusion and efficient decoupled

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning

Background:

• Existing YOLOv5s algorithm struggles with infrared target detection due to incomplete feature extraction and detection errors.
• Infrared target detection requires robust feature representation to overcome challenges like low resolution and thermal noise.

Purpose of the Study:

• To develop an improved infrared target detection algorithm, YOLO-B, addressing limitations of YOLOv5s.
• To enhance feature extraction, information fusion, and prediction accuracy for infrared targets.

Main Methods:

• Proposed CSPPF structure to expand the receptive field of the feature extraction network.
• Implemented Bifusion Neck for effective fusion of shallow and deep features.
• Utilized an efficient decoupled head for prediction and WIoUv3 loss for bounding box regression.

Main Results:

• Each proposed improvement individually demonstrated superior detection accuracy.
• YOLO-B achieved a 1.9% increase in accuracy, 7.3% in recall, 3.8% in mAP@0.5, and 4.6% in mAP@0.5:0.95 over YOLOv5s.
• YOLO-B outperformed YOLOv7 and YOLOv8s in parameter efficiency and detection accuracy.

Conclusions:

• The YOLO-B algorithm significantly improves infrared target detection performance.
• The integrated enhancements in feature extraction, fusion, and prediction contribute to the algorithm's effectiveness.
• YOLO-B presents a promising advancement for infrared target detection applications.