Rapid detection of enzyme activities in tomato leaves using visible-near infrared hyperspectral imaging integrated with image segmentation

Tomatoes (Solanum lycopersicum L.) are a globally consumed vegetable, rich in essential nutrients. Enzymatic activity in tomato leaves serves as a key indicator of plant health, stress response, and disease resistance. However, traditional enzyme measurement methods are destructive and time-consuming. This study employed hyperspectral imaging (HSI) combined with image segmentation to non-destructively assess three critical enzymes: peroxidase (POD), phenylalanine ammonia-lyase (PAL), and α-amylase (α-AMS). Hyperspectral images of 125 tomato leaf samples (370–1715 nm) were analyzed using a novel segmentation algorithm to extract reflectance data. Chemical assays provided reference enzyme levels (POD: 5.108 mU/g; PAL: 7.347 U/g; α-AMS: 35.856 U/g). Genetic algorithm (GA) and successive projection algorithm (SPA) optimized wavelength selection, with GA-PLS (partial least squares regression) emerging as the top-performing model. Three alternative prediction models were developed by combining GA-selected spectra with support vector machine (SVM), particle swarm optimization-backpropagation neural network (PSO-BP), and random forest (RF) algorithms. It was found that the GA-PLS model had better prediction performance for the three enzymes in the near infrared band of leaves with Rp(average coefficient of determination for the three enzymes) and RMSEP(average root means square error prediction of the three enzymes) of 0.873 and 1.489, respectively. Results confirmed that near-infrared leaf spectra provided superior enzyme prediction compared to stems. The proposed segmentation method improved accuracy over manual techniques. This study demonstrates HSI's potential for rapid, non-destructive phytochemical analysis, offering valuable applications in precision agriculture and food quality control.