基于迁移学习和ResNet50的遥感图像土地覆盖分类

Land Cover Classification of Remote Sensing Images Based on Transfer Learning and ResNet50

摘要:

遥感和人工智能的飞速发展为基于图像的土地覆盖自动分类提供了实现途径,本文通过修改深度残差网络（ResNet50）的输出层，并利用迁移学习将网络在ImageNet数据库上的预训练参数模型作为土地覆盖分类网络的初始参数模型，通过在遥感图像数据库上继续训练实现对工业区、森林、停车场等7种土地覆盖类型的分类。分类准确率在RSSCN7和NWPURESISC45数据库分别达到92.32%和99.29%。结果表明，基于迁移学习的ResNet50深度学习算法能够实现遥感图像的快速、有效、精确的土地覆盖分类识别。

Abstract:

Rapid development of remote sensing technology and artificial intelligence has provided an implementation approach for image-based automatic land cover classification. In this paper, through modifing the output layer of the deep residual network (ResNet50), by using transfer learning and the pre trained parameter model of the network on the ImageNet database as the initial parameter model of the land cover classification network, by further fine-tuning training on the remote sensing image database, the industrial zone, forest classification of 7 types of land cover, including parking lots have been realized. The classification accuracy in the RSSCN7 and NWPU RESISC45 databases can reach 92.32% and 99.29%, respectively. The experimental results show that the ResNet50 deep learning algorithm based on transfer learning can achieve fast, effective, and accurate land cover classification and recognition of remote sensing images