Research On Hyperspectral Rapid Detection Method Of Rice Canopy Chlorophyll Content In Cold Region

Rice is one of the important food crops in our country.Improving the yield and quality of rice is the goal pursued now.During the growth of rice,changes in the level of chlorophyll content can be used as an indicator of crop nutritional status and whether it is infested by pests and diseases.However,the existing chlorophyll content detection mainly uses destructive sampling,and the content is mainly detected in the laboratory,which is difficult to meet the requirements of rapid detection.Therefore,the use of non-imaging hyperspectral remote sensing technology can quickly and non-destructively detect the chlorophyll content of crops,and provide a reference for subsequent rice growth monitoring,fertilization and pesticide management.In this study,cold rice in the northeastern region was used as the research object.When topdressing is needed at the seedling stage of rice,non-imaging hyperspectral sensors were used to obtain rice canopy spectrum information,and destructive sampling was used to obtain rice chlorophyll content.First,the canopy hyperspectral data is processed with the first derivative to remove the background interference,and then the continuous projection algorithm and the dimensionality reduction processing of discrete wavelet decomposition are used to obtain the best feature band and the best feature.Finally,a rice chlorophyll content inversion model based on four algorithms of support vector machine(SVM),extreme learning machine(ELM),nuclear extreme learning machine(KELM),and gray wolf optimized nuclear extreme learning machine(GWO-KELM)is constructed.The main research contents and results of this paper are as follows:(1)The characteristics of rice canopy spectrum at seedling stage were analyzed,and the effect of different chlorophyll content on rice canopy spectrum was studied.The hyperspectral reflectance of rice canopy leaves decreased with the increase of chlorophyll content.Derivative processing is performed on the rice canopy spectrum to remove the interference of the water body background and obtain relatively pure rice leaf spectrum information.(2)The pre-processed hyperspectrum was screened by the continuous projection method with characteristic bands,and 5 best characteristic bands were extracted,respectively: 416 nm,424nm,517 nm,680nm,707nm;the characteristic extraction was performed by discrete wavelet decomposition,The feature collection and number obtained are 15.(3)The data results of continuous projection algorithm and discrete wavelet decomposition are used as the input of the model,and the inversion of cold rice chlorophyll content based on support vector machine,extreme learning machine,nuclear extreme learning machine and gray wolf optimized nuclear extreme learning machine is established model.The results show that the chlorophyll content inversion model with the results of the continuous projection algorithm as the model input and the gray wolf optimization kernel extreme learning machine as the inversion model has the best effect.In this study,the non-imaging hyperspectral information of the rice canopy was obtainedto establish an inversion model between hyperspectral information and chlorophyll content.Provide some theoretical support for the rapid detection of rice chlorophyll,scientific fertilization,and pest monitoring in the future.