Small Object Detection with YOLOv8 Algorithm Enhanced by MobileViTv3 and Wise-IoU

Small object detection is an important and challenging task in computer vision, with widespread applications in fields such as remote sensing, autonomous driving, and security. However, due to factors such as small size, indistinct features, and complex backgrounds of small objects, traditional convolutional neural network (CNN)-based object detection algorithms often struggle to effectively detect small objects. To address issues such as insufficient feature extraction, suboptimal anchor box matching, and unbalanced loss functions, an improved YOLOv8 algorithm is proposed. This algorithm enhances the detection accuracy of small objects by adding a lightweight visual transformer, MobileViTv3, which can enhance feature representation capabilities, and a Wise-IoU loss function that adaptively adjusts the ratio of positive and negative samples and regression coefficients. Results show that the improved YOLOv8 algorithm achieves an increase of 2.5%, 2.1%, and 2.9% in precision, recall, and mAP respectively on the public remote sensing dataset VisDrone2019 and 2.4%, 2.7%, and 3.5% on DOTA compared to the original algorithm. It can effectively improve the accuracy of small object detection in complex scenarios while meeting the requirements for detection speed.