Monitoring of spider mite on eggplant leaves using spectral information and artificial intelligence

Abstract

Two-spotted spider mite (TSSM - Tetranychus urticae) is a very harmful parasite for various types of crops. It damages plants either piercing the cells of the plants to suck the sap contained and weaving cobwebs to defend itself from predators. Through these two mechanisms plants endure cellular respiration problems, and in the worst cases are brought to death. Since the treatment is more effective the sooner the infestation is identified, an early infestation diagnosis is required, and currently such analysis is performed through visual inspection. This study aims to propose an innovative diagnostic system based on the analysis of multi- and hyperspectral images. In particular, starting from terrestrial and aerial surveys of leaves obtained by multispectral sensors, we wanted to identify radiometric parameters that would allow to discriminate leaves where the spider mite was present from the leaves where it was absent. Statistical analysis of the analyzed parameters showed that Blue, Green, Red and Red-Edge, and Normalized Difference Vegetation Index (NDVI), Green NDVI (GNDVI), and Red-Edge GNDVI (REGNDVI), showed strong correlations with the state of presence of spider mite on the leaves, and, although in no case were clear correlations, results encourage classification research based on Artificial Intelligence (AI). We implemented algorithms based on hierarchical clustering and k-Nearest Neighbors (kNN) for classification purposes, and the results obtained encourage the development of both the tested techniques and others such as Fuzzy Logic (FL) or Artificial Neural Networks (ANN).