Transform-based image feature extraction method and system, and computer equipment

The invention discloses a transform-based image feature extraction method and system and computer equipment, and the method comprises the steps: firstly dividing a feature map into local windows with the same size, enabling each local window to refer to the information of an image region, achieving the interaction between the local windows through a local window shortcut (LWS), enabling the global image information to communicate with each other, and enabling the local windows to communicate with each other; and diversified global image information is added to the features extracted by the high-resolution transformer. The image features extracted by the method are suitable for high-precision image classification, so that the precision of the classification result is higher. 本发明公开一种基于transformer的图像特征提取方法及系统、计算机设备，该方法将特征图先划分为大小相同的局部窗口，每个局部窗口都指代一块图像区域的信息，通过局部窗口捷径(LWS)实现局部窗口之间的交互、使全局图像信息相互交流，为高分辨率transformer提取的特征增加了多样化的全局图像信息。本发明提取的图像特征适用于高精度图像分类中，使得分类结果精度更高。