A Large-area Land Cover Classification Method Based on Multilayer Perception Neural Network

Abstract

A large-area land cover classification method based on a multi-layer perceptual neural network, involving the field of land surface remote sensing technology, solves the problem that the current land cover classification method is difficult to effectively reflect the spatio-temporal pattern and conversion law of large-scale land cover, and cannot achieve large-scale applications. To solve problems such as limitations, calculate the surface reflectance of multiple remote sensing images, and use the automatic geo-registration algorithm to spatially correspond the image to be classified with the image of the surface classification result set, and automatically generate a reflectance sample set through the unsupervised sample library. A training set of highly generalized spatial-spectral feature data suitable for the multi-layer perception algorithm was constructed, and the model training was completed. Based on the trained multi-layer perceptual neural network model, the images in this scale range are interpreted, and at the same time, local optimization is performed in combination with the semantic proximity optimization model to improve the confusion after classification. And use the multi-GPU process block interpretation and mosaic method to quickly complete the classification and mosaic of a map of land cover.

Description

technical field [0001] The invention relates to the technical field of land surface remote sensing, in particular to a deep learning-based automatic land cover classification method and system. Background technique [0002] Global land cover data is a key source of information for understanding the complex interactions between human activities and global change, and is a variable in some key climate change studies (Imaoka et al. 2010). The surface cover classification products can provide products for the natural resource supervision, land use type monitoring, planting structure monitoring, planting area statistics and other businesses of governments at all levels. [0003] my country has entered a stage of rapid development of high-resolution earth observation technology. The continuous breakthrough of domestic satellite hardware technology has increased the spatial resolution, temporal resolution and even spectral resolution of remote sensing data, and the amount of remote...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problems that the current land cover classification method is difficult to effectively reflect the spatio-temporal pattern and conversion rules of large-scale land cover, and cannot realize large-scale application, which leads to limited use, the present invention provides a large-area land cover classification based on multi-layer perceptual neural network method

Images