A lightweight small object detection model for UAV images based on deep semantic integration.

Area of Science:

• Computer Vision
• Deep Learning
• Object Detection

Background:

• Existing small object detection methods often use computationally expensive residual blocks with redundant information.
• This complexity hinders performance improvements, especially for detecting small objects.

Purpose of the Study:

• To develop an efficient and effective small object detection method.
• To optimize feature fusion and reduce computational overhead in object detection models.

Main Methods:

• Designed an improved Feature Pyramid Network (L-FPN) for optimized resource allocation.
• Proposed the BPD-YOLO detector incorporating Dual-phase Asymptotic Feature Fusion (DAFF) and Deep Spatial Pyramid Fusion (DSPF) modules.
• Implemented a Decoupled feature Extraction-semantic Integration (DEI) mechanism for adaptive feature extraction based on feature map resolution.

Main Results:

• On the VisDrone dataset, BPD-YOLO with L-FPN improved mAP50 by 2.8% and mAP50-95 by 1.4% compared to the YOLOv8n + p2 baseline.
• BPD-YOLO demonstrated superior high-resolution feature extraction on the TinyPerson dataset.
• The proposed methods significantly reduced computational costs while enhancing detection accuracy.

Conclusions:

• The L-FPN, DAFF, DSPF, and DEI mechanisms effectively address the challenges of small object detection.
• BPD-YOLO offers a promising solution for accurate and efficient small object detection with reduced computational demands.