SmartCitrus: An efficient deep learning approach for real-time detection and classification of citrus leaf diseases

Abstract

Bangladesh is a prominent citrus exporter. Annually, the country has been exporting citrus fruits to over 60 countries. Distinguishing various diseases affecting citrus leaves requires a significant investment of time, effort, and specialized knowledge. Consequently, it is essential to create an innovative method for detecting citrus diseases. In this study, we have devised a valuable methodology by employing CNN models to identify diseases in citrus leaves. By employing a distinctive ensemble strategy, we successfully trained the model using varying numbers of classes in each stage. In reality, it allowed us to utilize suitable varieties of leaves for various ailments. Furthermore, it has enhanced the rate at which models learn during the later stages. In addition, it has reduced the level of model intricacy in comparison to frequently employed ensemble models. The identification of plant diseases in the present study involved the utilization of leaf photographs and algorithms for segmentation and feature extraction. Ultimately, we have successfully attained a 96 % accuracy rate for each class, signifying a substantial potential for mitigating production losses.