Selection of the optimal bands of first-derivative fluorescence characteristics for leaf nitrogen concentration estimation

• Published in: Applied optics (2004) 2019-07, Vol.58 (21), p.5720
• Main Authors: Yang, Jian, Cheng, Yinjia, Du, Lin, Gong, Wei, Shi, Shuo, Sun, Jia, Chen, Biwu
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Back propagation; Crop growth; Excitation; Excitation spectra; Feature extraction; Laser induced fluorescence; Monitoring; Neural networks; Nondestructive testing; Optical properties; Principal components analysis
• ISSN: 1559-128X; 2155-3165
• DOI: 10.1364/AO.58.005720

Laser-induced fluorescence technology provides a nondestructive and rapid method for monitoring leaf nitrogen concentration (LNC) based on its optical characteristics. Crop growth status can be efficiently diagnosed and quality evaluated by monitoring LNC. In this study, the first-derivative fluorescence spectrum (FDFS) was proposed and calculated based on the fluorescence spectra excited by 355, 460, and 556 nm excitation lights for rice LNC estimation. Then, the performance of each band FDFS characteristics and the FDFS ratio for LNC estimation were comprehensively discussed using principal component analysis and backpropagation neural network (BPNN). We analyzed the number of FDFS characteristics' influence on the accuracy of LNC monitoring. Results showed that R does not clearly improve for the LNC monitoring based on the BPNN model when the number of extracted FDFS features exceeds 4 or 5. Therefore, the FDFS optimal band combination of different excitation light wavelengths mentioned was selected for LNC monitoring. The selected band combinations contained the majority of FDFS characteristics and could effectively be applied in monitoring LNC (for 355, 460, and 556 nm excitation lights, with R of 0.764, 0.625, and 0.738, respectively) based on the BPNN model.