Assessing the severity of cotton Verticillium wilt disease from in situ canopy images and spectra using convolutional neural networks

• DOI: 10.1016/j.cj.2022.12.002
• Publication Date: 2022-12-23T10:53:57Z
• Authors: Xiaoyan Kang; Changping Huang; Lifu Zhang; Mi Yang; Ze Zhang; Xin Lyu
• Keywords: Verticillium wilt

Verticillium wilt (VW) is a common soilborne disease of cotton. It occurs mainly in the seedling and boll-opening stages severely impairs yield quality fiber. Rapid accurate identification evaluation VW severity (VWS) forms basis field cotton control, which has great significance to production. Cotton VWS values are conventionally measured using in-field observations laboratory test diagnoses, require abundant time professional expertise. Remote proximal sensing imagery spectrometry have potential for this purpose. In study, we performed situ investigations at three experimental sites 2019 2021 collected values, images, spectra 361 canopies. To estimate canopy scale, developed two deep learning approaches that use images spectra, respectively. For imagery-based method, given high complexity environment, first transformed task healthy diseased leaf recognition scene classification then built scenes (CFS) dataset with over 1000 each scene-unit type. We pretrained convolutional neural networks (CNNs) training validation CFS used after classify units canopy. The results showed DarkNet-19 model achieved satisfactory performance estimation (R2 = 0.91, root-mean-square error (RMSE) 6.35%). spectroscopy-based designed one-dimensional regression network (1D CNN) four layers. After dimensionality reduction by sensitive-band selection principal component analysis, fitted 1D CNN varying numbers components (PCs). top 20 PCs best 0.93, RMSE 5.77%). These learning-driven offer assessing crop from spatial spectral perspectives.