高光谱成像
边距(机器学习)
计算机科学
预处理器
模式识别(心理学)
人工智能
冗余(工程)
选择(遗传算法)
生物系统
机器学习
生物
操作系统
作者
Tingting Wang,Guanghui Li,Chenglong Dai
标识
DOI:10.1016/j.infrared.2022.104119
摘要
As an effective non-destructive detection technology, hyperspectral imaging (HSI) is widely applied in evaluating the quality of fruit, such as soluble solids content (SSC) in apples, oranges, pears, and sugar content in grape berries, and internal bruising in blueberries. But due to the redundancy in hyperspectral data, the prediction performance significantly relies on the characteristic wavelength selection. Most previously published studies rarely simultaneously consider the correlations among different spectral bands and the extraction of characteristic bands from the original spectrum. To solve the problem, this study explores the application of hyperspectral technology to determine soluble solids content (SSC) in Korla pears. It focuses on reducing the hyperspectral data by applying a new effective wavelength selection method called Group sampling Margin Influence Analysis (GsMIA). It combines band-correlation and band-influence to enhance the selection of key bands. GsMIA contains three steps: grouping, margin influence analysis, and second selection. Combined with S-G smooth first derivative preprocessing method and support vector regression models, the proposed method can effectively yield good prediction performance with only 7.81 % of the original wavelengths, and the experimental results of comparison with several state-of-the-art methods on Korla fragrant pears SSC datasets further demonstrate the effectiveness and superiority of the proposed method.
科研通智能强力驱动
Strongly Powered by AbleSci AI