Rust Detection on Coffee Crops Using Artificial Intelligence and Vegetation Indices

This paper presents innovative Artificial Intelligence (AI) driven solutions to detect Coffee Leaf Rust (CLR) in El Salvador and Central America. The coffee leaf analysis was combined with image preprocessing techniques, including Vegetation Indices (VIs) and segmentation with tools such as Segment Anything Model (SAM) and Python's Rembg library. Several machine learning algorithms were implemented, such as Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) to pre-trained neural networks such as MobileNet, DenseNet, GoogleNet, ResNet, AlexNet and VGG, plus late and early hybrid combinations of MobileNet and DenseNet, as well as proprietary architectures based on Convolutional Neural Networks (CNN). Model evaluation was performed using standard metrics such as precision, recall, F1 Score and accuracy. MobileNet and DenseNet architectures achieved over 90 % performanced. Also, the models responses were analyzed using confusion matrices, training times and response latency. The findings underscore the importance of properly selecting and configuring classifier architectures for accurate identification was highlighted, DenseNet201 and MobileNet emerging as the most effective. Comparisons with exisitng research highlight the competitiveness of the developed models and preprocessing methods, identifying areas for future studies.