AUTHOR=Xu Lijia , Chen Yanjun , Wang Xiaohui , Chen Heng , Tang Zuoliang , Shi Xiaoshi , Chen Xinyuan , Wang Yuchao , Kang Zhilang , Zou Zhiyong , Huang Peng , He Yong , Yang Ning , Zhao Yongpeng 

TITLE=Non-destructive detection of kiwifruit soluble solid content based on hyperspectral and fluorescence spectral imaging

JOURNAL=Frontiers in Plant Science

VOLUME=Volume 13 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2022.1075929

DOI=10.3389/fpls.2022.1075929

ISSN=1664-462X

ABSTRACT=Soluble solid content (SSC) is one of the important parameters to determine the quality and taste of kiwifruit. This study explored the hyperspectral imaging technique and fluorescence spectral imaging technique combined with multiple regression models to predict SSC in kiwifruit, respectively. First, a total of 90 samples of kiwifruit was collected from a planting base in Sichuan province, divided into 70 calibration sets and 20 prediction sets, and the hyperspectral and fluorescence spectral images of the samples were collected and the regions of interest were extracted. Second, six methods, such as the standard normal variate transform, the detrend correction, Savitzky-Golay convolution smoothing, the gaussian window smoothing, the boxing smoothing, and the exponential smoothing, were used to preprocess the two spectral data, and the best preprocessing method was selected after comparing the prediction results. Then, the primary feature extraction algorithms, including the bootstrapping soft shrinkage (Boss), the competitive adaptive reweighted sampling (CARS), the iteratively variable subset optimization (IVSO), the interval variable iterative space shrinkage approach (IVISSA), the model adaptive space shrinkage (MASS), and the secondary feature extraction algorithm (i.e., CARS-Boss, MASS-Boss, IVISSA-Boss), were used to extract feature variables from the preprocessed spectral data. Subsequently, three regression prediction models, i.e., the extreme learning machines (ELM), the partial least squares regression (PLSR) and the particle swarm optimization - least square support vector machine (PSO-LSSVM), were established, respectively. We further analyzed and compared the model prediction results, which showed that the MASS-Boss-ELM based on fluorescence spectral imaging technique had the best prediction performance for the SSC of kiwifruit, with the corresponding "R" _"p" ^"2" , "R" _"c" ^"2"  and "RPD"  of 0.8894, 0.9429 and 2.88, respectively. The MASS-Boss-PLSR based on hyperspectral imaging technology had a slightly lower prediction performance, with the "R" _"p" ^"2" , "R" _"c" ^"2" , and "RPD"  of 0.8717, 0.8747, and 2.89, respectively. The results suggest that both spectral imaging techniques have good potential for application in the detection, and among them, the fluorescence spectral imaging technology is more conducive to improving the prediction performance, and provides theoretical support and technical means for non-destructive detection of fruit internal quality.