Citrus Diseases Detection using Deep Learning

Abstract

Pakistan is a major contributor to citrus fruit production, accounting for 30% of the total fruit output, with citrus cultivation spread across all four provinces, particularly Punjab. Citrus is vital to domestic and international markets and is distributed through various value chains. However, like many fruits, citrus is susceptible to diseases such as canker, citrus scab, and black spot, impacting fruit quality and quantity. Manual disease diagnosis in citrus fruits is time-consuming, error-prone, lacks standardization, and incurs high costs, requiring expert intervention. Accurate diagnosis and treatment are imperative for safeguarding citrus crops. The implementation of automated systems provides efficient, consistent, and cost-effective solutions, mitigating the challenges associated with manual diagnosis and contributing to sustainable citrus farming practices. This paper introduces an automated system employing Deep Learning and optimal feature selection for classifying citrus diseases. The process begins with data augmentation, enhancing the training dataset by creating additional images from existing samples. Two pre-existing models, DenseNet-201 and AlexNet, undergo adaptation and retraining utilizing an augmented dataset via transfer learning techniques. The experiment is carried out on the leaves dataset.Attaining the highest accuracy of 99.6%. The suggested framework is examined at every phase and compared to modern methods approaches, affirming its superior performance.