Deep learning satellite data cloud detection algorithm using unified sample

Abstract

The invention discloses a deep learning satellite data cloud detection algorithm using a unified sample. The method comprises the following steps: 1, establishing a hyperspectral data pixel library; 2, simulating multispectral data; 3, performing cloud detection based on a neural network; 4, performing result analysis. The method comprises, firstly, selecting enough cloud and clear sky pixels on different earth surface types based on AVIRIS data with high spatial hyperspectral resolution; simulating spectral reflectance of cloud and clear sky pixels of the to-be-identified satellite based on the spectral response function of the to-be-identified satellite data and the AVIRIS waveband width, and taking the spectral reflectance as an input vector of the neural network; and inputting the simulated reflectivity into a neural network model, training a neural network cloud detection model corresponding to the sensor, and carrying out cloud detection. And in combination with visual interpretation, the accuracy of cloud detection results of different underlying surfaces and different satellite sensors is compared and analyzed. According to the method, cloud pixel detection can be achieved,cloud detection of various multispectral sensors is supported, and the adaptability of cloud detection is wider.

Description

technical field [0001] The present invention relates to the use of a deep learning satellite data cloud detection algorithm using uniform samples. Background technique [0002] In optical remote sensing, the quality of satellite data has a great influence on the retrieval accuracy of remote sensing parameters, and cloud cover is a very important factor that dominates the quality of remote sensing data. According to the International Satellite Cloud Climatology Project, clouds usually exist on 66% of the earth's surface, and general satellite remote sensing platforms will be affected by clouds and cannot receive energy, thereby reducing the quality of satellite images. [0003] The electromagnetic radiation information received by the sensor includes the geometric distribution and geographic information of the ground objects, which can be used for urban planning, coastal zone detection, and agricultural production estimation. However, due to the cloud cover problem in the sa...

AI Technical Summary

Problems solved by technology

Studies have shown that the cloud detection method based on statistical learning is better than the threshold method in terms of accuracy and universality, but the method is more complicated and less efficient

Images