SENet-SVWF: a spectral vegetation indices and wavelet features fusion method using squeeze and excitation network for predicting the SPAD value of maize leaves.

Hyperspectral technology is used to monitor the SPAD (Soil and Plant Analysis Development) of maize, which is of great significance for regulating maize growth, optimizing nutrient management and improving yield formation. However, too much spectral information of hyperspectral data has resulted in the parameters redundancy and the complex structure of model. To solve this problem, a spectral vegetation indices (VIs) and wavelet features (WF) fusion method using squeeze-and-excitation network (SENet-SVWF) was proposed in this study. The partial least squares regression (PLSR), support vector regression (SVR), random forest regression (RFR), extreme gradient boosting (XGBoost) and long short-term memory (LSTM) predictors and five pretreatment methods were used in predicting SPAD value of maize, and the best pretreatment method and predictor were determined. VIs and wavelet transform (WT) were used to analyze the spectral data after the best pretreatment. Pearson correlation analysis and peak extraction method were used to determine the combination of VIs and WF. The SENet was introduced to fuse the selected VIs and WF, and the corresponding weights of different spectral features were optimized. The results showed that the multivariate scattering correction (MSC) was identified as the best pretreatment method, and SVR and RFR predictors performed best. The model established using SE-VIs-WF spectral features combined with RFR has the highest accuracy. The R