Glacier Area Monitoring Based on Deep Learning and Multi-sources Data
Glaciers are the source of fresh water that have obvious response on climate change. It is crucial and meaningful to monitor glacier area. In this paper, a glacier detection method based on features derived from low-resolution optical data, thermal data a
- PDF / 2,581,282 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 91 Downloads / 167 Views
2
Liaoning Technical University, Huludao 125105, China Quanzhou Institute of Equipment Manufacturing Haixi Institutes, Chinese Academy of Sciences, Quanzhou 362216, China [email protected]
Abstract. Glaciers are the source of fresh water that have obvious response on climate change. It is crucial and meaningful to monitor glacier area. In this paper, a glacier detection method based on features derived from low-resolution optical data, thermal data and TerraSAR-X images is proposed. The Land Surface Temperature (LST) was firstly obtained by Convolutional Neural Network (CNN). Combined with the velocity information, a low-resolution binary mask was derived for the supervised classification of SAR imagery. Afterwards, a set of suitable features was derived from the SAR intensity image, such as texture information generated based on the gray level co-occurrence matrix (GLCM), and intensity values. With these features above, the glaciers were classified by Random Forest (RF) to distinguish the glacier from the non-glacier areas. Compared to the unsupervised classification only using SAR data, the glacier detection method proposed in this paper achieved a better performance with the overall classification accuracy of 90.88%. Keywords: Glacier area
CNN LST Optical data Thermal data SAR
1 Introduction The study of glacier contributes to the understanding of the law of glacier change [1], which is one of the important research contents in the field of glacier change and global climate change [2]. LST can be used as a feature to distinguish the glacier from the surrounding environment, especially in summer. The traditional method assumes a simple linear relationship between surface temperature and brightness and emissivity, and then using the relationship to obtain LST retrieval results [3]. However, these methods simplify many preconditions, such as atmospheric constant correction, absorption by surface vegetation, and ignoring atmospheric scattering effects. These assumptions affect the accuracy of retrieval to some extent. The neural network algorithm has less dependence on physical equation of radiation transmission, which can solve the problem of ill-condition retrieval of physical model [4]. Neural network does not need to deduce the physical process for specific problems, only needs to input a representative data set containing multiple samples to train the neural network, and the trained neural network model is used for the retrieval © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 Q. Liu et al. (Eds.): CENet 2020, AISC 1274, pp. 409–418, 2021. https://doi.org/10.1007/978-981-15-8462-6_46
410
G. Wang et al.
[5]. In this paper, CNN is used to achieve the goal of LST retrieval. It is meaningful to introduce CNN to construct the relationship between Brightness Temperature (TB) and Land Surface Temperature (LST). In addition, Synthetic Aperture Radar (SAR) is widely applied to map glaciers because SAR is not affected by weather conditions,
Data Loading...
