A Small-Object-Detection Model Based on Improved YOLOv8 for UAV Aerial Images
Kaijie Niu, Yuxuan Yan
Abstract
Aiming at the problems of high model complexity and poor detection effect of UAV object detection algorithm, a small-object-detection model based on improved YOLOv8 is proposed. First, the Diverse Branch Block generalized module component is used in the Backbone network, which can improve the performance of the convolutional neural network in the model training phase and does not affect the inference time in the inference phase. The number of parameters of the model is reduced without decreasing the detection accuracy. Secondly, a kind of Asymptotic Feature Pyramid Network is introduced in the Neck network, which improves the small-object detection ability of the model. Finally, the Wise-IoU loss function is integrated to improve the overall performance of the detector. Experimental results on VisDrone, show that the improved model improves the mean average precision (MAP) by 9.5% and reduces the number of parameters by 37% compared to the baseline YOLOv8-s model. Overall, the improved algorithm has a large improvement in detection effect and can be applied and deployed to real-world scenarios.