Sensoriamento espectral assistido por aprendizado de máquina em plantas sob estresse abiótico e iluminação seletiva

Abstract

Nanotechnology has emerged as a promising approach in agriculture to optimize crop performance; however, nanoparticles can also be unintentionally introduced into the environment, resulting in effects that are still not fully understood in plant development. In this context, the development of reliable and non-destructive techniques capable of identifying plant exposure to nanoparticles and evaluating their physiological responses becomes increasingly important. This study investigates the combined use of reflectance spectroscopy and machine learning analysis to characterize leaves of tomato plants irrigated with different solutions (water, silver nanoparticles, and gold nanoparticles) and subjected to different lighting conditions (red, green, blue, and white). Reflectance spectra were acquired in the visible range (400–800 nm), enabling classification and regression analyses. Classical machine learning algorithms, including K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM), were used to classify the plants according to the irrigation solution and lighting condition applied to each sample. Among the evaluated models, the Support Vector Machine (SVM) achieved the best performance, reaching an F1-score above 0.96, with 2085 correctly classified samples out of 2160 in the test set, demonstrating the robustness of the spectral data for classifying leaves exposed to different nanoparticle solutions and lighting conditions. In addition to classification, chlorophyll analyses were performed, allowing the quantification of chlorophyll content in the plants and the evaluation of its relationship with the different irrigation treatments and lighting conditions. Overall, the results demonstrate that reflectance spectroscopy combined with machine learning techniques is an effective and non-destructive approach, capable of classifying plants subjected to different treatments (irrigation and lighting) and estimating their chlorophyll content. This methodology has the potential to become a powerful tool in precision agriculture, supporting plant development studies and contributing to increased crop productivity through optimized conditions and treatments.