Evaluating Maize Grain Quality by Continuous Wavelet Analysis Under Normal and Lodging Circumstances

Root lodging is the most common stress that occurred in maize growing period. It has a great impact on both yield and grain quality. This study aims at developing some leaf level non-contact detecting models of grain quality for both maize plants under lodging and normal circumstances. In Anthesis and Spinning stages of maize growth, an artificial lodging was manipulated to simulate the naturally occurred physical force like windstorm. The hyperspectral measurements of three-ear-leaves were taken for both normal and lodged plants by the ASD FieldSpec Pro spectrometer. The contents of oil, protein and starch in the grain were measured by an automated near-infrared grain analyzer. A two-tailed t-test was used to identify the grain quality properties with significant difference between normal and lodging treatments. A continuous wavelet analysis (CWT) was employed to extract the spectral features of grain quality contents. Based on these spectral features, a partial least squares (PLS) regression was applied in developing the predicting models of grain quality parameters for both normal and lodging samples. As shown in the results, the wavelet transformed spectral features were successfully generated for both protein and starch samples, yet with varied wavelength positions and decomposition scales between normal and lodging treatments. The accuracy of the multiple regression models was relatively high, with an R-2 over 0.75 for all predicting models. The potential of CWT analysis in predicting maize grain quality parameters under both normal and lodging circumstances was thus illustrated.