AIE-YOLO: Auxiliary Information Enhanced YOLO for Small Object Detection

Area of Science:

• Computer Vision
• Deep Learning
• Object Detection

Background:

• Small object detection is a significant challenge in computer vision.
• Existing models like YOLOv5 struggle with information loss during feature extraction for small objects.

Purpose of the Study:

• To improve the sensitivity and detection performance of YOLOv5 for small objects.
• To address information loss and enhance feature representation for small objects.

Main Methods:

• Proposed an auxiliary information-enhanced YOLOv5 model.
• Introduced a context enhancement module with multi-scale receptive fields and an attention branch.
• Integrated high- and low-frequency information via wavelet transform into the PANet for feature fusion.

Main Results:

• Achieved a 9.5% higher mean average precision compared to the original YOLOv5 on the Tsinghua-Tencent 100 K dataset.
• Maintained real-time detection speeds.
• Outperformed mainstream object detection models.

Conclusions:

• The proposed auxiliary information-enhanced YOLOv5 effectively improves small object detection.
• The integration of context enhancement and wavelet transform-based feature fusion is crucial for enhancing small object recognition.
• The model offers a promising solution for real-time small object detection in challenging datasets.