关键词: 苹果, 霉心病, 阻抗特性, 稀疏主元分析-支持向量机, 稀疏主元分析-人工神经网络

Abstract:

In order to explore and establish an non-destructive method for the detection of mouldy core in apples, the
changes of seven impedance parameters with frequency and three physiochemical quality properties were measured on
normal fruits and mouldy-core fruits of Fuji apple using LCR instrument under 100 Hz to 3.98 MHz, 1.0 voltage and
constant temperature of (20 ± 1) ℃. The results showed that with increasing frequency, complex impedance (Z) and
parallel resistance (Rp) of apples revealed a decrease whereas susceptance (B) and conductance (G) exhibited an increase.
Significant linear relationship between lgf and lgZ or lgB was observed, respectively (R2 ＞ 0.99); phase angle (θ), logarithm
of parallel capacitance (lgCp) and logarithm of loss coefficient (lgD) revealed a fluctuating trend, with 1, 2 and 3 turning
points, respectively. Mouldy-core incidence did not change the trend of each impedance parameter-frequency curve in fruits,
but the values of some parameters were altered, for example, Z decrease, B and Cp increased. Totally 27 specific impedance
parameters with non-zero loading coefficient for the construction of 14 effective principal components were screened
through sparse principal component analysis (SPCA). When recognizing mouldy-core fruits using classifiers of support
vector machine (SVM) and artificial neural network (ANN), SVM provided higher accuracy. In addition, in 10 crossvalidation
classification tests, mouldy core fruits could be discriminated with an accuracy of 94%, confirming the validity of
impedance parameters screened and the feasibility of SPCA-SVM analysis software for discriminating mouldy-core from
normal fruits. Moreover, the incidence of mouldy-core fruits presented a decrease in density and soluble solid contents in
comparison to normal fruits, which provided a physiochemical basis for changes in bio-impedance properties.

Key words: apple, mouldy core, impedance property, Sparse principal component-SVM, SPCA-ANN