Orange Disease Detection and Classification: A Multi-class Orange Fruit Dataset Creation and Deep Learning-based Classification Models

Oranges are among the most widely cultivated fruits in the world, contributing significantly to the agricultural economy and a balanced diet. However, citrus diseases are a major threat to orange production and cause substantial economic losses. The eye observation option, often used to identify these diseases, is time-consuming and error-prone. The contribution of this paper is two-fold. First, a new dataset is introduced, containing 9000 images divided into 3 classes: healthy and those affected by two orange diseases, citrus canker and citrus melanose, and it is made publicly available. To our knowledge, it is the most comprehensive data set on orange fruit disease detection. Second, the papers compare the performance of different deep learning-based models on the dataset, including MobileNetV2 and DenseNet121. These models achieved an average classification accuracy of 83.49% and 95.06%, respectively.