System and method for classifying information in image

The invention provides a method for classifying information in an image, which comprises the following steps: a convolutional neural network receives an input image and generates a plurality of shared feature maps, an attention network generates a plurality of attention maps according to the shared feature maps, a fusion circuit selects at least two attention maps from the attention maps to execute fusion operation so as to generate a fusion map, and the fusion map is used for generating the information in the image. And the classifier generates a classification result according to the fusion image. The architecture corresponding to the classification system provided by the invention can learn all related symptoms at the same time and is also specific enough, and each individual task can be evaluated. 本发明提供一种影像中信息的分类方法，包括：卷积神经网络接收一输入影像并产生多个共享特征图，注意力网络依据这些共享特征图产生多个注意力图，融合电路从这些注意力图中选择至少二个执行融合运算以产生融合图，以及分类器依据融合图产生分类结果。本发明提出的分类系统所对应的架构可以同时学习所有相关症状，并且也足够具体，可以评估每个单独的任务。