文档介绍：: .
激光与光电子2%、%，较
CBAM 模块提高了 %和 %。此外，为了将模型移植到移动端，结合迁移学****提出
了 MobileNet_v2+PA_CBAM 的识别方法，准确率达到 %，较之前提升 %。研究表
明，提出的注意力机制模块 PA_CBAM 在野生菌细粒度识别研究中具有更好的识别效果，
具有一定的泛用性。并且，MobileNet_v2+PA_CBAM 训练后模型大小仅为 ，识别图
片的平均耗时仅为 ，是在移动端部署野生菌识别的理想模型。
关键词 图像识别; 细粒度; 特征融合; 注意力机制; 迁移学****
中图分类号 TP 文献标志码 A
Fine-grained Classification Research of Wild Mushrooms Based on
Feature Fusion and Attention
Qian Jiaxin1, Yu Pengfei1*, Li Haiyan1, Li Hongsong1
1 School of Information Science and Engineering, Yunnan University, Kunming, Yunnan
650500, China
Abstract Identifying the species of wild mushrooms is an important way to prevent
poisoning by eating those toxic. Therefore，in order to improve the accuracy of the fine-grained
classification of wild mushrooms, a new attention module PA_CBAM improved from CBAM
changes the connection way of channel attention module and spatial attention module from
serial to parallel in CBAM, then add their results together. As a result, the problem of
interference caused by cascade of these two attention modules is solved. In addition, the
proposed method improves the ResNet50 by referring to the idea of feature pyramid, whose
accuracy of the Top-1 and Top-5 are % and %, which are % and % higher
than the original method. Furthermore, the Top-1 and Top-5 reach % an