A remote sensing data analysis method based on depth learning

Abstract

The invention discloses a remote sensing data analysis method based on depth learning, which can further improve classification accuracy and reduce the influence of the number of member models on classification accuracy. The method comprises the following steps of 1) reading the original data of the hyperspectral remote sensing data; 2) determining a marking sample; and selecting training samplesand test samples from the marked samples; 3) carrying out unsupervised self-coding depth network learning on all train samples; 4) marking samples for supervise in-depth network learning; 5) inputtingthat unlabeled samples into the depth network to obtain a classification result; 6) selecting the best sample from the unlabeled samples by using the active learning image classification method basedon the least difference sampling; 7) labeling the samples selected by active learning, deleting the samples from the unlabeled samples, adding the unlabeled samples into the training samples, updating the training samples, and obtaining the final classifier; 9) evaluating the accuracy of the classification results. By sampling the method, the classification accuracy and the classification efficiency can be improved.

Description

technical field [0001] The invention relates to the field of analysis of remote sensing data, in particular to a remote sensing data analysis method based on deep learning. Background technique [0002] The remote sensing data source is a series of ground object spectral data with different wavelengths obtained through active or passive remote sensors. After the obtained ground feature spectral data is converted into digital data, it can be used in geographic information system. [0003] Most Landsat scanners are typically spectral scanners, a linear array of devices. The scanner acquires ground data by reciprocating scanning (such as LANDSAT) or advancing scanning (such as SPOT). These scanners passively record the solar spectrum reflected from the earth's surface, such as multispectral scanning data can extract crop type, distribution and growth, topography, soil, water surface and river network information. Infrared scanners can record thermal radiation data emitted by...

AI Technical Summary

Problems solved by technology

[0005] However, the above method still has the shortcomings of low classification accuracy and the number of member models has a greater impact on classification accuracy.

Images