A Nondestructive Method Combined Chlorophyll Fluorescence With Visible-NIR Spectroscopy for Detecting the Severity of Heat Stress on Tomato Seedlings
Heat stress can inhibit the growth of tomato seedlings and lead to yield loss.Temperature is often used as an indicator to evaluate the impact of plant heat stress.However,due to the difference between individual plant heat tolerance and their health status,plants under the same temperature may show different heat stress symptoms,which could lead to misclassification.Therefore,combined with chlorophyll fluorescence technology and visible near-infrared spectroscopy,this paper proposes a classification method for tomato seedlings'heat stress severity.The chlorophyll fluorescence parameters and visible near-infrared(Vis-NIR)spectrum data of the controlled and heat-stressed plants were collected.Using multiple chlorophyll fluorescence parameters as indicators,a clustering model based on the k-means++algorithm was established to obtain the classification labels on the severity of heat stress.The reasonableness of the clustering result was verified by analyzing the difference between the chlorophyll fluorescence parameters and the biochemical indicators among the three samples.Then,the spectral data were labelled based on the output of the clustering model;six characteristic bands highly related to the sample's heat-stress severity were extracted using three preprocessing methods and their combinations,combined with three characteristic wavelength selection algorithms.With six characteristic bands as input and the heat-stress-severity as output,classification models are established based on four machine learning algorithms to classify the heat-stress-severity.The results showed that The chlorophyll fluorescence parameters Fv/Fm,Fv/Fo,NPQ,Y(NPQ)and Y(NO)showed significant moderate to high correlation with their heat stress status,and the samples were labelled as non-heat-stressed samples,mild heat-stressed samples and severe heat-stressed samples based on the five parameters.After feature extraction,more than 99％of redundant features are eliminated,and only six characteristic wavelengths remain.Characteristic wavelengths that can be used to establish classification models are obtained.The LDA model performs best among the four models,with a classification accuracy of 92.45％,an F1 score of 0.929 1,and an AUC of 0.978 0.The results indicate that using chlorophyll fluorescence technology combined with Vis-NIR technology to detect heat stress is feasible.This study provides an effective method for rapidly detecting heat stress,rapid screening of heat tolerance in plants and early warning of heat stress.
Heat stressChlorophyll fluorescencek-means++algorithmVisible-near-infrared spectroscopyClassification model
万方数据
维普
