Classification of Chili Fruit Diseases Using Deep Convolutional Neural Network Transfer Learning

Abstract

Chili peppers are among the highest-value agricultural commodities, often experiencing significant price fluctuations due to supply constraints. The rainy season frequently leads to crop failures caused by diseases affecting chili plants. Existing methods often struggle to accurately differentiate between similar symptoms on leaves and fruits, leading to misdiagnosis and ineffective disease management strategies. Early detection of these diseases, which manifest as symptoms on the leaves and fruits, is crucial for effective pest management. Common diseases include anthracnose, characterized by dry brown spots on the fruit, and fruit rot, where the interior of the fruit decays while the skin remains intact. Identifying these diseases promptly is essential for applying appropriate treatments to ensure optimal yields.In this study, a comprehensive approach is taken to classify diseases in chili pepper plants (Capsicum annuum L.) by incorporating both leaf and fruit segmentation. The research employs Deep Convolutional Neural Networks with Transfer Learning (DCNN) to enhance detection capabilities. The findings reveal that for leaf disease classification, fewer neurons in additional layers yield better accuracy and reduced loss, while for fruit disease classification, a more complex model with additional neurons is necessary. This underscores the need for balancing model complexity to achieve optimal performance and prevent overfitting, particularly in distinguishing between leaf and fruit diseases.