Abstract:
Object detection on Unmanned Aerial Vehicles (UAVs) is a challenging problem due to the limited computing resources of the edge GPU of the Internet of Things (IoT) nodes ...Show MoreMetadata
Abstract:
Object detection on Unmanned Aerial Vehicles (UAVs) is a challenging problem due to the limited computing resources of the edge GPU of the Internet of Things (IoT) nodes and the presence of a large number of small objects in aerial images. Therefore, this paper proposes a lightweight deep learning architecture based on YOLOX model. Firstly, we design a lightweight backbone network to replace the backbone network in YOLOX. Then, we use four different sizes of neck feature maps for detection, which can improve the accuracy of small object detection much better. At the same time, we reduce the number of parameters by removing one convolution from the header and adding a max-pooling layer to obtain local information for classification. Compared to YOLOX-s, our model has improved the mAP@50 and mAP@0.5:0.95 by 4.6% and 2.5% respectively on the Visdrone2023 validation set. It is worth noting that our model has only 6.61M parameters, and we also provide a tiny version with only 2.59M parameters. A series of experimental results illustrates that our enhanced algorithm outperforms YOLOX-s, YOLOV7-tiny, and the latest YOLOV8-s.
Published in: 2024 27th International Conference on Computer Supported Cooperative Work in Design (CSCWD)
Date of Conference: 08-10 May 2024
Date Added to IEEE Xplore: 10 July 2024
ISBN Information: