Classification Of Peanut And Prediction Of Moisture And Protein Content In Peanut Using Hyperspectral Imaging

Peanut is a significant source of edible vegetable oil and protein, the content of constituents such as moisture and protein plays an important role in peanut quality and the peanut storing and processing, which varies greatly in different varieties of peanut. Therefore, discriminating peanut varieties and detecting protein quality is a crucial procedure in the peanut production and selling. Most of the traditional detection methods are not only costly and time-consuming but also need to destruct the integrity of peanut samples, hyperspectral imaging, integrated computer vision and spectroscopy, has been emerged to overcome these shortcomings. This study was conducted for classification of peanut varieties and prediction of moisture and protein content, the following are details:(1) Visible and near-infrared hyperspectral imaging (400-1000nm) coupled with classifiers was conducted to discriminate five peanut varieties. Seven classifiers were used to establish discrimination models. PLS-DA, LDA, and PNN using full wavelengths showed the CCR of calibration set of 100%, the corresponding CCR of prediction set was 93.33%, 94.44%, and 71.11%, respectively. In addition, six optimal wavelengths (416,518,572,633, 746, and 928 nm) were designated by SPA and used to simplify the classification models. The simplified model obtained by the PNN classifier also presented good prediction accuracy and the CCR of calibration set was 83.81%, demonstrating that the integration of hyperspectral imaging and classifiers analysis was feasible to discriminate peanut varieties.(2) Calibration models were created using spectral data and moisture content values by PLSR technique in tandem with spectral pre-processing techniques. The performance of two spectral ranges of 400-1000nm and 1000-2500nm was compared for further moisture content analysis of peanut. The original PLSR prediction model, established using the full spectral wavelengths in the spectral range of 400-1000nm showed the best prediction ability with the highest R2P of 0.930 and the lowest RMSEP of 0.054%. In addition, optimal wavelengths in the spectral range of 400-1000nm that carried the most influential moisture content information were selected by Regression coefficients (RC) and SPA to solve the collinear problems and remove the redundant information. Moreover, the performance of PLSR models established using optimal wavelengths selected using the above-mentioned two variable selection methods were compared. Finally, the optimised RC-MLR model yielded the best results with R2p of 0.937 and RMSEP of 0.051%, which was used to visualize the moisture content distribution in peanut, revealing that vis-near hyperspectral imaging (400-1000nm) as an promising tool for measurement of moisture content distribution of peanut.(3) The potential of long-wave near-infrared hyperspectral imaging technique (1000-2500 nm) for prediction of protein content in peanut was investigated. The PLSR prediction model without any spectral pre-processing, established using the full spectral wavelengths showed the best prediction ability with R2p of 0.885. Regression coefficients and successive projection algorithm were utilized to recognize the most important wavelengths that possessed the greatest influence on the protein content prediction based on the whole spectral range. By comparing two variable selection methods, eight (1153,1567,1972,2143, 2288,2339,2389 and 2446 nm) optimal wavelengths were selected by RC and its corresponding simplified prediction model was obtained, showing R2p of 0.870. Moreover, the quantitative MLR calibration model showed the best result, with R2p of 0.878, revealing the potential of hyperspectral imaging for prediction of protein content in peanut.