深度学习在遥感图像地类识别中的研究与应用

Research and Apllication of Remote Sensing Image Recognition on Land Sort with Deep Learning

以深度学习卷积神经网络为基础,采取迁移学习的方法,开展对遥感影像林地、灌木林地、草地、耕地和城镇等5大地类的自动识别研究。通过对Inception-V3网络进行合理的优化改进,训练后得到适合遥感影像地类判别的训练模型。应用该模型,对内蒙古鄂尔多斯市准格尔召地区的遥感影像进行了自动识别与分类,并获得了较为理想的地类识别效果。

Based on deep learning CNN,the research of remote sensing image recognition on land sort which contained forest,shrub,meadow,farmland and urban land,is carried out by transfer learning.The model used in remote sensing image recognition is given after training by optimized and improved Inception-V3 net.Using trained model to recognize and classify remote sensing image on Junggar district,Ordos,the result of recognition on land sort is satisfactory.