Neural network implementation for a reversal procedure for water and dry matter estimation on plant leaves using selected LED wavelengths

Elian Conejo (1) , Jean-Pierre Frangi (2) , Gilles de Rosny (2)

1 UCR - Universidad de Costa Rica
2 IPGP - Institut de Physique du Globe de Paris

An inversion method based on a neural network was used to estimate water and dry matter contents on plant leaves, from transmittance and reflectance measurements, using light emitting diodes (LEDs) at specific wavelengths in NIR and FIR. The preliminary results for the predicted water content by the neural network method showed a RMSE value of 0.0027 g/cm(2) and vertical bar sigma vertical bar value of approximately 3.53%, computed on 127 plant leaf samples over 51 species. Dry matter estimation also was performed, which showed potential implementation after future improvements. We believe this inversion method could be implemented in a portable system based on any silicon platform with the capability to perform in situ measurements on plant tissue. (C) 2015 Optical Society of America

Keywords: KeyWords Plus:LEAF OPTICAL-PROPERTIES; MODEL; DIAGNOSIS; PROSPECT; STRESS; SENSOR; WHEAT

Domains: Sciences of the Universe [physics]; Earth Sciences

Elian Conejo, Jean-Pierre Frangi, Gilles de Rosny. Neural network implementation for a reversal procedure for water and dry matter estimation on plant leaves using selected LED wavelengths. Applied optics, 2015, 54 (17), pp.5453. ⟨10.1364/AO.54.005453⟩. ⟨insu-02924968⟩