Prediction of antioxidant enzyme activity in tomato leaves based on microhyperspectral imaging technique

Abstract

Antioxidant enzymes play an important role in defending of tomato leaves against damage under different types of adversity stresses. In this paper, hyperspectral imaging (HSI) technology was used to predict the peroxidase (POD), catalase (CAT) and superoxide dismutase (SOD) activities in tomato leaves. Spectra was preprocessed using Baseline, Smoothing and Normalize methods. Characteristic wavelengths were selected using competitive adaptive reweighted sampling (CARS), interval random frog (IRF) and genetic algorithm-combined partial least squares (GAPLS). Macroscopic prediction models for antioxidant enzyme activity were developed using partial least squares regression (PLSR), principal components regression (PCR) and least square support vector machines (LSSVM) algorithms. To establish microscopic prediction models for microzone cell image and spectral detection, the antioxidant enzyme activity in the observed leaf area was migrated to the microzone using the macroscopic prediction model and the ratio of microzone area to sliced leaf area. As results of model evaluation, the correlation coefficients of calibration (Rc) and prediction (Rp) were 0.853 and 0.832 for POD activity, 0.790 and 0.715 for CAT activity, and 0.975 and 0.937 for SOD activity, respectively. The results indicated the possibility of transfer learning between macroscopic and microscopic models in quantitatively predicting antioxidant enzymes in tomato leaves by HSI. The findings in this study provided a reference data in quantitatively detection of trace substances in tomato leaves and suggested macroscopic to microscopic modelling approaches for the future HSI applications.