Detection of Water Content in Lettuce Canopies Based on Hyperspectral Imaging Technology under Outdoor Conditions

Abstract

To solve the problem of non-destructive crop water content of detection under outdoor conditions, we propose a method to predict lettuce canopy water content by collecting outdoor hyperspectral images of potted lettuce plants and combining spectral analysis techniques and model training methods. Firstly, background noise was removed by correlation segmentation, proposed in this paper, whereby light intensity correction is performed on the segmented lettuce canopy images. We then chose the first derivative combined with mean centering (MC) to preprocess the raw spectral data. Hereafter, feature bands were screened by a combination of Monte Carlo uninformative variable elimination (MCUVE) and competitive adaptive reweighting sampling (CARS) to eliminate redundant information. Finally, a lettuce canopy moisture prediction model was constructed by combining partial least squares (PLS). The correlation coefficient between model predicted and measured values was used as the main model performance evaluation index, and the modeling set correlation coefficient Rc was 82.71%, while the prediction set correlation coefficient RP was 84.67%. The water content of each lettuce canopy pixel was calculated by the constructed model, and the visualized lettuce water distribution map was generated by pseudo-color image processing, which finally revealed a visualization of the water content of the lettuce canopy leaves under outdoor conditions. This study extends the hyperspectral image prediction possibilities of lettuce canopy water content under outdoor conditions.