Multi-task neural network framework for remote sensing scene classification and classification method

Abstract

The invention relates to a neural network framework for remote sensing scene classification and a classification method, in particular to a multi-task neural network framework for remote sensing sceneclassification and a classification method, and solves the problems of limitation of information amount, inaccurate scene recognition and low classification precision of existing network frameworks and classification methods. The network framework comprises a convolution feature extraction layer, a classification task full-connection feature extraction layer, a classification task discriminationlayer and a classification task loss layer; the network framework is characterized by further comprising an auxiliary task full-connection feature extraction layer, an auxiliary task discrimination layer, an auxiliary task loss layer, a classification task feature mapping layer, an auxiliary task feature mapping layer and a relationship learning loss layer. Wherein the two feature mapping layers respectively carry out dimensionality reduction on full-connection feature vectors adapted to two tasks, the relation learning loss layer carries out subtraction on the vectors after dimensionality reduction and takes norms of difference vectors as relation learning losses, and the relation learning losses and discrimination losses of the two tasks are added into optimization training together.

Description

technical field [0001] The invention relates to a neural network framework and a classification method for remote sensing scene classification, in particular to a multi-task neural network framework and classification method for remote sensing scene classification. Background technique [0002] The goal of the high-score remote sensing scene classification task is to distinguish the foreground of interest from the background and other foregrounds according to the information of the input image. [0003] Scene classification methods can be divided into two types based on manual features and based on deep learning. The method based on manual features uses artificially designed features. In the past period of time, this method has achieved good results; but with the rise of deep learning, the disadvantages of manual feature-based methods relying on human factors have been revealed, and depth Learning can independently learn favorable features, which greatly improves the averag...

AI Technical Summary

Problems solved by technology

[0010] The purpose of the present invention is to provide a multi-task neural network framework and classification method for remote sensing scene classification, to solve the existing neural network framework and classification method for remote sensing scene classification, due to limitations in the amount of information, scene recognition is not Accurate, technical issues with low classification accuracy

Images