Nondestructive Detection Of Apples’ Quality And Varieties By Dielectric Spectra

The production and consumption of fruit in China has been in the front rank of the world. However, the quality control for fruits in China had not reached the international level. The question restricted the development of fruits exports. Therefore, some noval techniques are demanded to detect the fruit internal quality accurately, rapidly and comprehensively.Based on the dielectric property of apples, the analysis models of detecting soluble solids content(SSC), acidity(pH), moisture content(MC), identifying apples variety were established with the experimental materials of Fuji, Red rome and Pink lady. The principal component analysis(PCA), uninformative variables elimination method(UVE), based on partial least squares, and successive projection algorithm(SPA) were applied to extract characteristic variables from original full dielectric spectra. The generalized regression neural network(GRNN), learning vector quantization(LVQ) network, support vector machine(SVM) and extreme learning machine(ELM) modeling methods were used to establish models to predict SSC, pH, MC and varieties of apples, based on the original full dielectric spectra and characteristic variables, respectively. The main results and conclusions are shown as follows:(1) Sample sets were partitioned by set partitioning based on joint X-Y distances(SPXY), based on the original dielectric spectra, the model for detecting SSC of Fuji apples established by ELM coupled with SPA had a relatively optimal prediction precision and feasibility. The correlation coeffiencient of calibration(Rc), root mean square error of calibration(RMSEC), correlation coeffiencient of prediction(Rp), root mean square error of prediction(RMSEP) of the model were 0.904, 0.681, 0.892 and 0.605 respectively.(2) Sample sets were partitioned by SPXY, based on the original dielectric spectra, the model for detecting pH of Fuji apples established by ELM coupled with SPA had a relatively optimal prediction precision and feasibility. The Rc, RMSEC, Rp, RMSEP of the model were 0.741, 0.087, 0.720 and 0.078 respectively.(3) Sample sets were partitioned by SPXY, based on the original dielectric spectra, the model for detecting MC of Fuji apples established by ELM coupled with SPA had a relatively optimal prediction precision and feasibility. The Rc, RMSEC, Rp, RMSEP of the model were 0.926, 0.606, 0.910 and 0.609 respectively.(4) Sample sets were partitioned by Kennard-Stone method, based on the original dielectric spectra, the model for detecting varieties of apples established by ELM coupled with SPA had a relatively optimal prediction precision and feasibility. The accrcay is 99.6% for all prediction apple samples.