改进的MobileNetV2遥感图像场景分类方法

Improved remote sensing image scene classification method based on MobileNetV2

为了解决传统方法在遥感图像场景分类中存在的低准确率和低精度问题,本文提出一种基于迁移学习与注意力机制的遥感图像场景分类方法。首先,以第二代移动网络(MobileNetV2)为基础网络模型;其次,引入迁移学习防止模型出现过拟合的现象;最后,引入注意力机制模块使模型关注图像中关键的特征信息。在航空图像数据集(AID)、遥感图像场景数据集(NWPU45)进行了实验,准确率分别达到了96.88%、95.45%。实验结果表明,本文方法可以有效地提高遥感图像场景的分类精度。

In order to solve the problems of low accuracy and low precision of traditional methods in remote sensing image scene classification,a remote sensing image scene classification method based on transfer learning and attention mechanism was proposed.Firstly,the second-generation mobile network(MobileNetV2)was used as the base network model;secondly,transfer learning was introduced to prevent the model from overfitting;finally,the attention mechanism module was introduced to make the model focus on the key feature information in the image.Experiments were conducted on the aerial image dataset(AID)and remote sensing image scene dataset(NWPU45),and the accuracies reached 96.88%and 95.45%,respectively.Experimental results show that the proposed method in this paper can effectively improve the classification accuracy of remote sensing image scenes.