A Lightweight Model of Multi-UAVs Object Detection with Dynamic Layer Aggregation

Aiming at the problems of missing detection, false detection and poor real-time performance in the UAV small object detection of aerial images under the complex background of dim, dense trees and buildings, we propose a lightweight YOLO small object detection algorithm with dynamic layer aggregation. Firstly, we propose a dynamic efficient layer aggregation lightweight Backbone, and gradient path planning is adopted to improve the feature extraction ability for small targets. Secondly, we propose a dynamic Neck with Omni-Dimensional dynamic convolution. It helps obtain the variability of each dimension and richer context information, and reduce the missing rate of the model for small targets. Thirdly, SIoU loss function is designed to improve the detection accuracy of small targets and accelerate the convergence of the model. Finally, the channel pruning operation is carried out to compress the size of the model without affecting the performance of the model. Multi-UAVs detection dataset was constructed to evaluate the performance of the proposed model. Compared with the baseline YOLOv5, [email protected] of the proposed model improved by 49.2%, which means its good performance in detecting small targets under complex background environments. And FPS reaches 78.236, meeting the requirements of real-time detection.