Investigation of Nitrogen/Potassium deficiency in Alternanthera sessilis plant using deep learning model combined with CF-LIBS approach

Abstract

In this paper, the investigation on Nitrogen (N)/Potassium (K) deficiency in Alternanthera sessilis plant is carried out by using a deep learning model combined with Calibration free – Laser induced breakdown spectroscopy (CF-LIBS) technique. The trained deep learning model considered is ResNet50 and DenseNet201. The models are trained and evaluated on a custom created dataset with three categories, namely Healthy, Nitrogen deficit, and Potassium deficit. A prediction accuracy of 92.10% and 98.89% is obtained using ResNet50 and DenseNet201 models, respectively. The obtained results are validated using the CF-LIBS technique. In LIBS, a high energy pulsed Nd:YAG laser (wavelength = 1064 nm, 532 nm and 355 nm) with a pulse duration and repetition rate of 6 ns and 10 Hz is used to create ablation and generate plasma. Laser irradiance varied between 1 × 10¹⁰ W/cm² to 3 × 10¹⁰ W/cm². LIBS data is used to determine the nutrient content. After extensive experimental investigation, the estimated nutrient concentration for a healthy leaf sample is Ca = 6795 ± 645 ppm, N = 4284 ± 572 ppm and K = 14407 ± 609 ppm. The samples with Nitrogen (N = 1178 ± 541 ppm) and Potassium (K = 8989 ± 581 ppm) deficits show a reduction in respective concentrations. The determined Nitrogen (N)/Potassium (K) deficiency in Alternanthera sessilis plant using a deep learning model and CF-LIBS method are related to each other. The specified method can be used to perform in-situ analysis, rapid, remote measurement and multielement identification/ranking of plant materials.