基于改进EfficientNet的细粒度图像识别

Fine-grained Image Recognition Based on Improved EfficientNet

普通CNN模型直接应用于细粒度图像识别时关键特征提取不充分,导致模型细粒度识别准确率较低,针对这个问题,论文提出了一种基于改进EfficientNet的细粒度图像识别算法,以EffcientNetB3为主干,在全局平均池化层(GAP Layer)之前添加一个CBAM注意力模块,提升模型关键特征提取能力。论文利用迁移学习训练得到细粒度识别网络,实验结果表明,训练得到的改进模型在CUB-200-2011数据集上的识别准确率达到了84.5%左右,相比于原网络准确率提升了5.4%,另外与常用CNN模型相比模型复杂度更低,识别准确度更好。

When the ordinary CNN model is directly applied to fine-grained image recognition,the key features are not extracted sufficiently,resulting in low recognition accuracy.In order to solve this problem,this paper proposes a fine-grained image recognition algorithm based on improved EfficientNet,with EffcientNetB3 as the backbone,and adds a CBAM attention module before the Global Average Pooling Layer to improve the key feature extraction ability of the model.In this paper,the fine-grained recognition network is trained by transfer learning,and the experimental results show that the recognition accuracy of the improved model on the CUB-200-2011 dataset reaches about 84.5%,which is 5.4%higher than that of the original network,and the model complexity is lower and the recognition accuracy is better than that of the commonly used CNN model.