Deep Learning-Based Monitoring of Chili Plant Leaf Health

Abstract

Chili is one of the most important crops in Sri Lanka, but its productivity and quality are often threatened by leaves diseases and nutrient deficiencies such as Bacterial Spot, Cercospora Leaf Spot, Curl Virus, White Spot, and general Nutritional Deficiencies. Traditional identification methods, including leaf testing or relying on expert advice, are slow, expensive, and not easily accessible to new chili farmers in rural areas. This study proposes a two-stage chili leaf analysis system based on Convolutional Neural Networks (CNNs). In stage 1, a CNN model with image preprocessing and data augmentation techniques (rotation, flipping, zooming, brightness adjustment, rescaling) separates chili leaves from non chili leaves, achieving a training accuracy of 99.97% and a validation accuracy of 99.96%. This step ensures that only chili leaf images are used for further analysis. In the second stage, another CNN model with deep learning architectures, batch normalization, and optimization algorithms classifies chili leaves into six categories: bacterial spot, cercospora leaf spot, curl virus, white spot, nutritional deficiency, and healthy leaves. This model achieved 92.33% training accuracy and 89.42% validation accuracy, demonstrating its effectiveness in detecting leaf diseases and nutrient deficiencies. Once fully deployed, the proposed system can provide farmers with a fast, practical, and cost-effective solution to monitor the health of their chili crop, support timely decision-making, and minimize potential yield losses.