Exploration of the earth environment using “Himawari-8” data of meteorological satellite and deep learning

บทคัดย่อ

In the field of satellite remote sensing, artificial intelligence (AI) is the method frequently used to explore the global environment in many researches. Deep learning is widely used to analyze the satellite images. The detection or tracking on typhoons, clouds, land and snow are examples of research on the earth’s environment with deep learning. The conventional method, RGB and IR images of the satellites, are used as the inputs of deep learning using convolution operations such as convolution neural network (CNN). We have developed the method to investigate the earth’s environment and its variations. The method is based on the supervised learning model of artificial intelligence with 16 wavelength spectral image data (from visible to infrared range) obtained by the geostationary meteorological satellite "Himawari-8". The input layer was given information of 16 wavelengths instead of an image, not convolution operation. The output from the proposed algorithm is classification of 6 classes such as typhoon, cloud, sea, land, snow and ice. We also succeeded in visualizing that typhoons and developed clouds contain snow and ice. The combination of spectrum data acquired by satellites and our proposed method is effective for forecasting typhoons, and drift ice, torrential rain and so on. By providing spectral data of the meteorological satellite “Himawari-8” to the input layer of the proposed method, our model can recognize the earth environment in all space and time (365 days, 24 hours). The model developed this time has reached the accuracy of 96.5%. It will be an effective method for drawing out the capabilities of artificial satellites in the environment exploration of the earth and planets.