Hyperspectral Image Database Query Based on Big Data Analysis Technology

Beixun Qi,F Wen,S.M Ziaei

doi:10.1051/e3sconf/202127503018

Abstract

In this paper, we extract spectral image features from a hyperspectral image database, and use big data technology to classify spectra hierarchically, to achieve the purpose of efficient database matching. In this paper, the LDMGI (local discriminant models and global integration) algorithm and big data branch definition algorithm are used to classify the features of the hyperspectral image and save the extracted feature data. Hierarchical color similarity is used to match the spectrum. By clustering colors, spectral information can be stored as chain nodes in the database, which can improve the efficiency of hyperspectral image database queries. The experimental results show that the hyperspectral images of color hyperspectral images are highly consistent and indistinguishable, and need to be processed by the machine learning algorithm. Different pretreatment methods have little influence on the identification accuracy of the LDMGI model, and the combined pretreatment has better identification accuracy. The average classification accuracy of the LDMGI model training set is 95.62%, the average classification accuracy of cross-validation is 94.36%, and the average classification accuracy of the test set is 89.62%. Therefore, using big data analysis technology to process spectral features in hyperspectral image databases can improve query efficiency and more accurate query results.

Highlights

Deep learning has made significant progress in the hyperspectral image classification task, but manual labeling of hyperspectral images is still time-consuming and laborious, and deep learning methods usually rely on a large number of manually labeled samples, when only a small number of labeled samples, the deep learning model is difficult to train
Using the deep learning method for hyperspectral image classification is still faced with the problem of lack of labeled training samples and deep neural network training
Spectral information can be stored as chain nodes in the database, which can improve the efficiency of a hyperspectral image database query

Summary

Introduction

Deep learning has made significant progress in the hyperspectral image classification task, but manual labeling of hyperspectral images is still time-consuming and laborious, and deep learning methods usually rely on a large number of manually labeled samples, when only a small number of labeled samples, the deep learning model is difficult to train. Liy proposed a geometric-based color theme extraction method, which uses convex hull in RGB color space to represent a color theme, and transforms the problem of color theme extraction into the problem of convex hull generation and simplification in geometric space [4]. On this basis, fan h further proposes an improved method based on iterative optimization, which improves the problem of poor color representation in color themes [5]

Methods

Results

Conclusion