Using UAV image data to monitor the effects of different nitrogen application rates on tea quality. (9th September 2021)

• Record Type: Journal Article
• Authors: Luo, Danni; Gao, Yuan; Wang, Yu; Shi, Yujie; Chen, Sizhou; Ding, Zhaotang; Fan, Kai
• Is Part Of: Journal of the science of food and agriculture. Volume 102:Number 4(2022)
• Page Start: 1540
• Page End: 1549
• Publication Date: 2021-09-09
• Subjects: unmanned aerial vehicle (UAV) -- remote sensing -- nitrogen -- partial least squares regression (PLSR) -- support vector machine (SVM) -- BP neural network; Food -- Periodicals; Agriculture -- Periodicals; 664
• DOI: 10.1002/jsfa.11489
• Languages: English
• ISSNs: 0022-5142

Abstract

Abstract: BACKGROUND: Accurate and efficient evaluation of the effect of nitrogen application rate on tea quality is of great significance for nitrogen management in a tea garden. However, previous methods were all through soil or leaf sampling, using biochemical methods for laboratory testing. These methods are not only less one‐time detection samples, but also time‐consuming, laborious and inefficient. Therefore, the development of fast, efficient and non‐destructive diagnostic methods is an important goal in this field. RESULTS: We obtained spectral information on the tea canopy using a multispectral camera carried by an unmanned aerial vehicle (UAV), and extracted the average DN value of the experimental plot by environmental visual imagery (ENVI); we finally obtained 28 spectral parameters. By analyzing the correlation between spectral parameters and ground parameters measured synchronously, five spectral parameters with high correlation were selected. Finally, the prediction models of tea nitrogen, polyphenol and amino acid content were established by using support vector machine (SVM), partial least squares and backpropagation neural network. Through modeling comparison and coefficient verification, the results show that the ground parameters measured in the laboratory were in good agreement with the results estimated by the model. The SVM model had the best performance in predicting nitrogen and tea polyphenol content, with R 2 = 0.7583 and 0.7533, root mean square error of prediction (RMSEP) = 0.4086 and 0.3392, and normalized RMSEP (NRMSEP) = 1.23 and 1.28, respectively. The partial least squares regression model had the best performance in predicting amino acid content, with R 2 = 0.7597, RMSEP = 0.1176 and NRMSEP = 4.10. CONCLUSION: The results show that the model based on UAV image data and machine learning algorithm can effectively detect the main biochemical components of the tea plant, which provides an important basis for tea garden management. © 2021 Society of Chemical Industry.