Research On Prediction Methods Of Pre-market Potato Germination Stage Using Hyperspectral Imaging Technology

Detecting the germination stage of potatoes is an important part of defect detection. Germination is a complex biological process, predicting the germination stage timely can reduce the loss caused by germination. In this research, prediction of the pre-market germination stage was conducted using hyperspectral imaging technology. The main contents are as follows:(1) A software based on hyperspectral imaging technology to predict the germination stage of potatoes was developed by using the MATLAB platform, including hyperspectral image processing module and data analysis module. Hyperspectral image processing module consisted of quick actions of images, the operation of ROI and batch processing of images. This module was able to extract the location parameter of each batch potato eyes semi-automatically, assuring that each potato could obtain the hyperspectral images during the period of germination stage for two weeks at the same potato eyes' area. The data analysis module included processing spectral data, extracting feature parameter and predicting model analysis. The parameters of Sine Fit Algorithm (SFA) and Characteristic Triangle Algorithm (CTA) were included in the parameters extraction of the software. Fisher Discriminant Model was used to classify the potato eyes according to the characteristic parameter extracted above.(2) SFA and CTA to obtain characteristic parameters to predict the germination stage of potatoes were proposed. SFA used sine functions combination of different amplitudes, frequencies and phases to fit the discrete spectral points to establish the function relationship between spectral values and wavelengths. Feature points in CTA were obtained by first derivative extreme value of spectra curve after using SFA, and using the intersection point of the extreme point and its tangent to construct the characteristic triangle. The parameters of CTA included cos B?S?H et.al, and SFA included 15 parameters. Both methods reduced the spectral data of the full bands.(3) Building prediction models based on CTA, SFA and SPA to predict the germination stage of potatoes. Compared with the discriminant rate of Fisher Discriminant Model and LSSVM Model under combinations of different characteristic parameters, we discovered that the discriminant rate of SPA-SFA-LSSVM model was the highest for potato eyes of apical buds, which was 80.65%. For other potato eyes, the discriminant rate of SPA-SFA-Fisher model was the highest, which was 97.6%.