Semi-supervised hyperspectral data analysis method based on double-flow conditional generative adversarial network

Abstract

The invention relates to a semi-supervised hyperspectral data analysis method based on a double-flow conditional generative adversarial network. The method comprises the following steps: S1, acquiringhyperspectral sample data; S2, performing image segmentation, and embedding each sample into a three-dimensional data block; S3, dividing a training sample set and a prediction sample set, performingdata preprocessing, and the prediction sample set also serves as a label-free sample to be added into training; S4, constructing an adversarial generative network based on a double-flow condition, the system comprises a generator network with a double-flow structure, a discriminator network and a regression device network, and is used for extracting and fusing spectral features and spatial topography features, the generator network is used for generating a sample, the discriminator network is used for discriminating the authenticity of the sample, and the regression device network is used forregression calculation of a quantitative analysis regression target value; S5, constructing loss functions of the generator network, the discriminator network and the regression device network; S6, training an adversarial generative network; and S7, obtaining a quantitative analysis regression target value of the prediction sample set by adopting the trained adversarial generative network.

Description

technical field [0001] The invention relates to the technical field of hyperspectral detection, in particular to a semi-supervised hyperspectral data analysis method based on a two-stream conditional confrontation generation network. Background technique [0002] Hyperspectral-based quantitative analysis technology has a wide range of application scenarios, including food adulteration detection, fruit sugar content detection, microbial content detection, organic matter content detection, etc. [0003] At present, the commonly used quantitative analysis methods of hyperspectral data include partial least squares regression (PLSR), least squares support vector machine (LS-SVM), multiple linear regression (MLR), etc. [0004] The inventor found in the research that there are two main problems in the current hyperspectral quantitative analysis methods: (1) most of the quantitative analysis methods use the average spectrum to model, such as PLS and LS-SVM, which fail to make full...

AI Technical Summary

Problems solved by technology

[0004] The inventor found in the research that there are two main problems in the current hyperspectral quantitative analysis methods: (1) most of the quantitative analysis methods use the average spectrum to model, such as PLS and LS-SVM, which fail to make full use of the hyperspectral The spatial information contained in the spectral data, how to effectively extract and fuse the spectral features and spatial morphology features in the hyperspectral data is an urgent need for further research.

(2) Existing hyperspectral data analysis methods, in actual application scenarios, the number of training samples is limited, and the accuracy and generalization ability of the model are poor

Images