An Improved Object Detection Algorithm Based on YOLOv10

Object detection is still an important research topic in computer vision. For years, YOLO series object detection framework has been the object of research and improvement by many researchers since its appearance. In this paper, the improved space-to-deepth downsampling convolution algorithm is used to replace the convolution downsampling algorithm of YOLOv10 aiming at the shortcoming that the downsampling convolution process in the original detection algorithm can not retain more spatial information. We introduced RepCSP module to improve C2f module in YOLOv10. By mixing up linear attention mechanism, the feature extraction module and linear attention mechanism are integrated, and the feature extraction ability of the model is improved. Combined with the improved convolutional downsampling network, the improved C 2fRepCSP module and the integrated linear attention mechanism, the model is trained and evaluated on VOC 2007 and VOC 2012 data sets, and the results show that the detection accuracy of the model has been improved.