Hyperspectral Monitoring of Powdery Mildew Disease Severity in Wheat Based on Machine Learning

• DOI: 10.3389/fpls.2022.828454
• Publication Date: 2022-03-22T21:46:21Z
• Authors: Zi-Heng Feng; Lu-Yuan Wang; Zhe-Qing Yang; Yan-Yan Zhang; Xiao Li; Li Song; Li He; Jian-Zhao Duan; Wei Feng
• Keywords: 2. Zero hunger; feature band selection; machine learning; Plant culture; 0401 agriculture, forestry, and fisheries; Plant Science; 04 agricultural and veterinary sciences; remote sensing monitoring; wheat powdery mildew; spectral transformation; SB1-1110

Powdery mildew has a negative impact on wheat growth and restricts yield formation. Therefore, accurate monitoring of the disease is great significance for prevention control powdery to protect world food security. The canopy spectral reflectance was obtained using ground feature hyperspectrometer during flowering filling periods wheat, then Savitzky-Golay method used smooth measured data, as original reflectivity (OR). Firstly, OR spectrally transformed mean centralization (MC), multivariate scattering correction (MSC), standard normal variate transform (SNV) methods. Secondly, bands above four data were extracted through combination Competitive Adaptive Reweighted Sampling (CARS) Successive Projections Algorithm (SPA) algorithms. Finally, partial least square regression (PLSR), support vector (SVR), random forest (RFR) construct an optimal model index (mean index, mDI). results showed that after Pearson correlation, two-band optimization combinations machine learning modeling comparisons, comprehensive performance MC spectrum best, it better pretreating data. combined with CARS-SPA algorithm able extract characteristic more effectively. number screened than by band positions evenly distributed. In comparison different methods, RFR performed best (coefficient determination, R 2 = 0.741-0.852), while SVR PLSR models similarly (R 0.733-0.836). Taken together, estimation accuracy transformation (MC-RFR) highest, 0.849-0.852, root error (RMSE) absolute (MAE) ranged from 2.084 2.177 1.684 1.777, respectively. Compared (OR-RFR), increased 14.39%, RMSE MAE decreased 23.9 27.87%. Also, stage grain stage, which due relative stability structure in stage. It can be seen without changing shape curve, use preprocess CARS SPA algorithms bands, methods enhance synergy between multiple variables, established (MC-CARS-SPA-RFR) covariant relationship disease, thereby improving mildew. research this study provide ideas realizing high-precision remote sensing crop status.