CONVOLUTIONAL NEURAL NETWORK CLASSIFIER FOR IDENTIFYING DISEASES OF AVOCADO FRUIT (Persea americana Mill.) FROM DIGITAL IMAGES

Abstract

Mexico is the main avocado (Persea americana Mill.) producer in the world; however, this fruit is susceptible to phytopathogenic fungi that reduce quality during postharvest. Therefore, the timely identification of these organisms using non-invasive tools, such as artificial intelligence analysis, is of interest to reduce economic losses. The objective of this study was to implement a machine learning model with a database of digital images of fruits collected in the field, by creating a convolutional neural network (CNN) classifier, training and validating it to identify healthy avocado cv. Fuerte fruits and fruits infected with scab (Sphaceloma perseae Jenkins) or anthracnose (Colletotrichum spp.) from the production area in the State of Morelos, Mexico. Healthy avocado fruits were collected in the field, as well as fruits with the scab and anthracnose from different orchards to generate a set of 569 digital images. Transformations of these increased the dataset to 3983 images. The CNN model was trained with a random partition of 80% of the images and validated for prediction with the remaining 20%. CNN achieved an overall correct classification accuracy of 80% with the validation set. In addition, the classifier was evaluated with a test set of 100 images not included in the original set and achieved an overall correct classification accuracy of 87%. The CNN deep learning classifier implemented in this study represents the feasible use of artificial intelligence for application in avocado disease identification from digital images at the postharvest stage.