Avocado Leaf Disease Detection Using Deep Learning

Abstract

Avocado is a food crop of the tropical and sub-tropical zone scientifically known as Persea americana due to its creamy texture and health-enhancing fats. Detecting diseases in avocado leaves is essential for maintaining tree health and ensuring optimal fruit production. This research focuses on the application of deep learning approaches in detecting diseases in leaves of avocado with the general purpose of designing a reliable diagnostic system that can assist in diagnosing diseases within early phases of their manifestation, hence effectively allowing farmers to manage their crops. The dataset was increased to 3600 images with 600 images of avocado leaves infected with Persia mites, Miners, Margin burn, and healthy leaves with the help of augmentation. Five models were evaluated: three proprietary deep learning models, namely, Convolutional Neural Network (CNN), and three others which are Xception, VGG19, InceptionV3, and MobileNetv2. The models were then trained on the leaves’ images to categories them into the four categories. The assessment of the models revealed generally satisfactory results; MobileNetV2 was in the lead with 98.52% accuracy; next came Xception at 94.63%; InceptionV3 at 93.80%; VGG19 at 91.02%; and then the custom CNN at 84.44%. Due to its high accuracy, MobileNetV2 is ideal for implementation in real-time urgent like on fields. The analysis shows how the deep learning models can be applied in an effort to help farmers to timely notice the diseased plants, hence improve on the health and thus the yields of the crops. The further development will include the implementation of these models to more accessible interfaces and the improvement of their stability through the acquisition of more data and learning from it constantly.