【摘要】 提出一种基于剩余平均池化注意力机制的并联特征融合的脑肿瘤图像分类方法，使用加入剩余平均池化注意力机制的融合反向残差结构提取更精确的浅层信息方法，使模型更关注与目标特征相关的信息；选取三个模块的特征信息分别经过条带池化处理后拼接，融合浅层和深层的特征信息后进行分类。该方法将脑肿瘤分为神经胶质瘤、脑膜瘤、垂体瘤三类，提升分类准确率，帮助医生完成脑肿瘤病人的早期诊断。更多还原

【Abstract】 Propose a brain tumor image classification method that integrates parallel feature fusion with a residual average pooling attention mechanism,using the fusion inverse residual structure with the addition of the residual average pooling attention mechanism to extract the shadow information with the accurate shallow information,so that more attention is paid by the model to the information related to the target features;the feature information of the three modules are selected to be spliced after strip pooling process respectively,and the classification is performed after fusing the shallow and deep feature information.The brain tumors are classified into three categories,i.e.,glioma,meningioma,and pituitary tumor by the method,which improves the classification accuracy and helps doctors to complete the early diagnosis of brain tumor patients.更多还原