IDENTIFICATION OF DISEASE IN TOMATO LEAVES USING MACHINE LEARNING CLASSIFIERS AND DIGITAL IMAGES

Resumen

Early identification of diseases in crops improves agronomic decision-making and has a positive impact on agricultural production. In this study, we evaluated three machine learning classifiers to identify three diseases in a tomato crop (Solanum lycopersicum) using chromatic characteristics of digital images of leaves, and a computational tool was developed for its practical use. The classifiers were support vector machine (SVM), multilayer perceptron (MLP), and histogram gradient boosting (HGB). The target classes were tomato yellow leaf curl virus (V), the fungus Septoria lycopersici (H), the acarid Tetranychus urticae (A), and healthy leaves (S). The images were preprocessed to eliminate anomalies and the selection algorithm by region was used to obtain pixels of representative color for each target class. The pixels were then transformed from RGB to the HSV color model to create the training database, which consisted of threecolor characteristics (H,S and V) and the associated target class. The three classifiers achieved similar prediction performance. According to the Kruskal Wallis test, there were no significant differences (p-value = 0.5117). SVM obtained an overall accuracy (Acc) of 93.3 %, MLP obtained a value of 93.2 %, and HGB of 93.1 %. Moreover, in performance at the class level (diseases), SVM obtained a higher F1 = 96 % in identification of symptoms caused by Septoria lycopersici and a lower F1 = 90 % in identification of symptoms caused by Tetranychus urticae. The computational tool developed, IDENTO v1.0, facilitated identification of the three leaf diseases in tomato based on optimized classifiers and constitutes an option for promoting the use of artificial intelligencein agriculture.