基于迁移学习的小样本细粒度图像分类方法

Small Sample Fine-grained Image Classification Method Based on Transfer Learning

在图像分类领域,由于细粒度图像具有类间差距小、类内差距大的特点,导致分类困难。细粒度图像往往存在训练数据不足的问题,使用大型卷积神经网络进行分类容易出现过拟合现象,导致出现分类网络庞大、网络推理时间长且精度不高的问题。为提高小样本细粒度图像分类的精度和速度,采用轻量化卷积神经网络结合迁移学习的方式训练细粒度图像分类网络,并使用随机数据增强方法扩充数据集。在CUB200鸟类图片数据集上的实验结果表明,该方法能够有效提升分类效率和准确率。

In the field of image classification,it is difficult to classify fine-grained images because of the small gap between classes and the large gap within classes.Fine grained images often have the problem of insufficient training data.Using large convolutional neural network for classification is prone to over fitting,which leads to large classification network,long timefortraining and low accuracy.In order to improve the accuracy and speed of small sample fine-grained image classification,the lightweight convolutional neural network combined with transfer learning is used to train the fine-grained image classification network.Random data enhancement method is used to expand the data set.Experimental results on cub200 bird image data set show that this method can effectively improve the classification efficiency and accuracy.