CAFE-YOLO: an object detection algorithm from UAV perspective fusing channel attention and fine-grained feature...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Object detection in drone imagery faces challenges like small objects, complex backgrounds, and poor lighting.
• These factors degrade feature representation and detection accuracy.

Purpose of the Study:

• To propose a novel object detection algorithm, CAFE-YOLO, to address the challenges in aerial imagery.
• To enhance the detection of small, occluded, and complexly-situated objects in drone footage.

Main Methods:

• Incorporated a channel attention mechanism into the backbone network to focus on critical features.
• Introduced a fine-grained feature enhancement module for local detail extraction.
• Designed a lightweight attention-guided feature fusion strategy in the detection head.

Main Results:

• The CAFE-YOLO algorithm demonstrated significantly improved detection performance on the VisDrone2019 dataset.
• Achieved a mean average precision (mAP) of 44.6% at an IoU threshold of 0.5.
• Showcased substantial improvements in overall detection accuracy and robustness in complex scenarios.

Conclusions:

• CAFE-YOLO effectively addresses key challenges in drone-based object detection.
• The algorithm offers a lightweight yet robust solution for aerial imagery analysis.
• Results indicate superior performance compared to existing advanced algorithms in complex environments.