Improved detection of soluble solid and anthocyanin in litchi fruits by normalized VNIR-SWIR transmittance
Teng Long1, Kangning Wei1, Binshan Huang1
1College of Electronic Engineering/College of Artificial Intelligence, South China Agricultural University, Guangzhou 510642, China; Guangdong laboratory for Lingnan Modern Agriculture, South China Agricultural University, Guangzhou 510642, China; National Center for International Collaboration Research on Precision Agricultural Aviationdc Pesticides Spraying Technology, South China Agricultural University, Guangzhou 510642, China; Guangdong Engineering Technology Research Center of Smart Agriculture, South China Agricultural University, Guangzhou 510642, China; State Key Laboratory of Green Pesticides, Guizhou University, Guizhou 550025, China.
Abstract:
Detection accuracy of internal component contents in fruits by hyperspectral imaging (HSI) suffered from the geometric structure and the nonlinear relation between the content and spectral features. These issues were respectively addressed by developing approaches based on spectral normalization and spectral features (SPF)-image features (SSF)-geometric structure features (GSF)-nonlinear features (NLF) fusing. For this purpose, VNIR-SWIR transmission HSI combined with partial least squares regression (PLSR) model was employed to detect the soluble solid content (SSC) and anthocyanin content (AC) in litchi fruits. It was revealed that spectral normalization combined with SPF-SSF-GSF-NLF fusing improved Rp2 of PLSR model for SSC and AC by 17.47 % and 11.85 %, and the values reached 0.9148 and 0.8455, respectively. Furthermore, litchi grading approaches based on the predicted SSC and AC were developed with a high classification accuracy of 95.17 %. These results demonstrated that the proposed approach was effective in improving the detection accuracy of litchi fruit quality.
