Hyperspectral image characterization and modeling for prediction of ipomeamarone content in sweet potato

In order to rapidly and accurately detect ipomeamarone using hyperspectral image technology sweet potato inoculated with Ceratocystis fimbriata (that can infect sweet potato with black spot disease, leading to the generation of ipomeamarone) was used as the research object, methods of hyperspectral characterization information extraction and robust model construction for ipomeamarone content were explored. After eliminating the effects of noisy signals and offsetting baseline, hyperspectral characterization information of ipomeamarone was extracted by the successive projection algorithm (SPA) and the competitive adaptive reweighted sampling algorithm (CARS), 26 and 48 optimal characteristic wavelengths were extracted, respectively; Based on the extracted characteristic wavelengths, back propagation neural network (BPNN), least squares support vector machine (LSSVM), radial basis function neural network(RBFNN), partial least squares regression(PLSR)and extreme learning machine (ELM) prediction models were established, respectively. The performance of prediction models was compared and the results showed that the LSSVM model based on characterization information extracted by CARS (CARS-LSSVM) was optimal, with a R 2 of 0.968 and a RMSE of 0.2678. the content change and distribution of ipomeamarone in sweet potato samples were clearly illustrated in pseudo-color plots based on the prediction results of the CARS-LSSVM mode. Simultaneously electron microscope scanning (SEM) was performed for the sweet potato samples, the SEM results showed that the texture of sweet potato was varied with the development of black spot disease. The research provided a new method for the detection of ipomeamarone; and provided a basis for the early prevention and control of sweet potato black spot disease.