基于空间遥感影像的居民地自动解译

Automatic Interpretation of Residential Areas Based on Space Remote Sensing Images

针对当前地物要素信息采集劳动强度大、智能化程度不高、效率低等技术瓶颈,本文以深度学习理论为基础,在Caffe框架上依托Digits网络服务器构建居民地数据集进行分类识别训练,建立样本数据集并完成模型训练工作。整合居民地复杂数据信息,设计了针对遥感影像自动解译居民地的操作流程。通过对实验结果分析,利用深度学习获取的数据模型可以实现地物在不同场景中的特征信息描述,能够较好地完成对居民地的轮廓以及细节提取,实现对居民地以及其他地物种类的分类,成功检测居民地的概率极大,对其他地物种类误判概率较小,检测结果初步达到了预期效果。实验证明:网络结构合理,图像识别精度较为理想,具有较强的鲁棒性。

In view of the current technical bottleneck problems of high labor intensity,low degree of intelligence and low efficiency in the collection of surface feature information,based on the deep learning theory,relying on the digits network server on the Caffe framework,this paper constructs the residential area data set for classification and recognition training,establishes the sample data set and completes the model training.Integrating the complex data information of residential area,the operation process of automatic interpretation of residential area based on remote sensing images is designed.Through the analysis of the experimental results,the data model obtained by deep learning can realize the feature information description of the ground objects in different scenes,can better complete the extraction of the outlines and details of the residential areas,and realize the classification of residential areas and other types of ground objects.The probability of successful detection of residential areas is very big,and the probability of misclassification of other types of ground objects is small,and the detection results are preliminarily achieved as expected.Experiments show that the network structure is reasonable,the image recognition accuracy is ideal,and it has strong robustness.