An improved small target detection algorithm was proposed for YOLOv7-tiny algorithm.The algorithm consisted of three main design points.The MobileViT block module was used to improve the feature extraction capability.The feature fusion performance was optimized based on the EVC Block module.The MPDIoU loss function was used instead of the CIoU loss func-tion,which was used to cope with the situation where the prediction frame and the real target frame had the same aspect ratio while the real sizes were different.Experimental results show that the mAP value of the improved algorithm is 42.5％on the VisDrone dataset,which is a 5.6％improvement compared to that of YOLOv7-tiny.When the input image size is 640×640 pixels,the improved FPS value is 39.5,which can meet the real-time detection requirements of UAVs on edge devices.
关键词
小目标检测/移动视觉变换器/航拍数据集/注意力机制/增强值通道块/多阶段交并比/卷积
Key words
small target detection/MobileViT/aerial dataset/Transformer/EVC Block/MPDIoU/CNN
维普
万方数据