Remote sensing image classification method based on pruning compressed neural network

Abstract

The invention discloses a remote sensing image classification method based on a pruning compressed neural network, and the method comprises the steps: training an initial neural network model for a remote sensing image recognition problem, and enabling the model to serve as a to-be-pruned initial model; calculating a sensitivity matrix of parameters in the model layer by layer by utilizing the sensitivity of an output function learned by the initial model to tiny change of the parameters; sorting the values of the sensitivity matrixes of the parameters, and trimming off unimportant parameters;retraining the residual weight which is not pruned; after one layer is trimmed, repeating the steps on the trimming process of the next layer. and classifying the remote sensing images by using the trimmed neural network model. The method provided by the invention is more accurate, practical and reliable to calculate the importance degree of each parameter in the model, so that the unimportant parameters are eliminated, and finally, a satisfactory compression ratio is obtained and is used for more efficiently classifying and identifying the remote sensing images.

Description

technical field [0001] The invention relates to the technical field of remote sensing image processing and recognition, in particular to a remote sensing image classification method based on a pruned compression neural network. Background technique [0002] In recent years, deep neural networks have made major breakthroughs in the fields of remote sensing target recognition and remote sensing image classification. However, although the performance of deep learning models is very powerful, the existing deep learning models have huge parameters and complex network structures. Therefore, It will bring difficulties in both computing and storage, and it is difficult to deploy to mobile devices or embedded devices with limited memory and computing resources. Studies have shown that deep learning models have serious over-parameterization problems. Not all parameters play a role in the model. Some parameters have limited functions, redundant expressions, and even reduce the performa...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the object of the present invention is to provide a remote sensing image classification method based on pruning and compression neural network, which can effectively solve the problem of huge parameter quantity and high computational complexity in the process of remote sensing image classification by deep neural network model. The problem makes the pruned and compressed neural network model more efficient in the classification of remote sensing images

Images