基于深度学习的竞技运动图像分割的应用

A method of Competitive Sports Image Segmentation Based on Deep Learning

文章提出了一种基于金字塔池化模型(Pyramid Scene Parseing Network,PSPNet)神经网络模型的竞技运动图像分割方法。其使用了一种具有速度快、精度高、易部署优点的卷积神经网络架构,利用残差网络实现了数据增强技术,进一步提高了分割的准确性和稳定性。实验结果表明,提出的改进后的PSP Net模型在图像分割中的准确率为94.59%,优于其他方法。该方法有望在竞技运动领域的视频图像分割任务中得到广泛应用。

The article proposes a neural network model based on the Pyramid Scene Parseing Network(PSPNet)for athletic sports image segmentation.It uses a convolutional neural network architecture with the advantages of fast speed,high accuracy and easy deployment,and implements a data enhancement technique using a residual network to further improve the accuracy and stability of segmentation.Experimental results show that the accuracy of the proposed improved PSP Net model in image segmentation is 94.59%,which is better than other methods.The method is expected to be widely used in video image segmentation tasks in the field of competitive sports.