An Improved Lightweight Real-Time Detection Algorithm Based on the Edge Computing Platform for UAV Images

• DOI: 10.3390/electronics12102274
• Publication Date: 2023-05-18T10:32:58Z
• Authors: Lijia Cao; Pinde Song; Yongchao Wang; Yang Yang; Baoyu Peng
• Keywords: FLOPS; Minimum bounding box

Unmanned aerial vehicle (UAV) image detection algorithms are critical in performing military countermeasures and disaster search rescue. The state-of-the-art object algorithm known as you only look once (YOLO) is widely used for detecting UAV images. However, it faces challenges such high floating-point operations (FLOPs), redundant parameters, slow inference speed, poor performance small objects. To address the above issues, an improved, lightweight, real-time was proposed based on edge computing platform In presented method, MobileNetV3 YOLOv5 backbone network to reduce numbers of parameters FLOPs. enhance feature extraction ability MobileNetV3, efficient channel attention (ECA) mechanism introduced into MobileNetV3. Furthermore, order improve objects, extra prediction head neck structure, two kinds structures with different parameter scales were designed meet requirements embedded devices. Finally, FocalEIoU loss function accelerate bounding box regression localization accuracy algorithm. validate improved algorithm, we compared our other VisDrone-Det2021 dataset. results showed that YOLOv5s, MELF-YOLOv5-S achieved a 51.4% reduction number 38.6% decrease MELF-YOLOv5-L had 87.4% 47.4% fewer FLOPs, respectively, higher than YOLOv5l.