Urban impervious bed extraction method based on deep belief network

Abstract

The invention discloses an urban impervious bed extraction method based on a deep belief network, and the method comprises the steps: 100, obtaining the registered full polarimetric SAR data and multispectral image data; 101, carrying out the preprocessing of the full polarimetric SAR data to obtain the polarization features of the preprocessed full polarimetric SAR data; 102, carrying out the preprocessing of the multispectral image data to obtain the optical features of the preprocessed multispectral image data; 103, combining the polarization features and the optical features into a featurevector; 104, inputting the feature vector into a trained classifier to obtain a ground object classification result; 105, combining the classes of ground objects in the ground object classification result according to the ground object classification result to obtain an impervious bed distribution map, wherein the classifier is obtained by pre-training the deep belief network through a training sample. The method provided by the invention improves the classification accuracy of ground objects, and effectively improves the extraction precision of the urban impervious bed.

Description

technical field [0001] The invention belongs to remote sensing image processing technology, in particular to a method for extracting urban impermeable layers based on a deep belief network. Background technique [0002] Urbanization is one of the concentrated manifestations of intense human activities changing nature. One of the concentrated manifestations of urbanization is the increase of a large number of impervious surfaces, and the impact of urbanization on the environment is the impact of a large number of impervious surfaces on the environment. . Although the treatment of various sewage, waste gas and solid waste has maintained a standard level, the continuous rapid urban development is bound to greatly increase the pressure of various pollution treatment. In addition, various pollutants accumulated in the city enter the atmospheric circulation and The water cycle has caused non-point source pollution of the air and water in the entire city, which is even more diffic...

AI Technical Summary

Problems solved by technology

However, when the weights are randomly initialized, the magnitude of the gradient (from the output layer to the initial layer of the network) decreases sharply as the depth of the network increases, and the gradient of the overall loss function relative to the weights of the first few layers is very small, that is, the first layer The weights change so slowly that deep neural networks cannot effectively learn from samples, a problem often referred to as "gradient dispersion".

Images