The traditional YOLOv8 model has some limitations in oral image recognition,such as low image resolution and in-accurate localization of dental lesion regions.To address the above problems,the original YOLOv8 model is improved by improving the original YOLOv8 model,i.e.,adding the adaptive spatial correlation pyramid attention mechanism(ASCPA)to the backbone structure of the original model and replacing part of the ordinary convolution module(Conv)with the full-dimensional dynamic con-volution(ODconv).Secondly,the activation function of the original model is then optimized.The experimental results show that the improved YOLOv8 model has improved the mean accuracy(mAP)and recall of dental disease recognition by 3.3 percentage point,respectively,which provides a strong support for further research in dental medical image processing.
万方数据