Identifying Multiple Diseases on a Single Citrus Leaf Using Deep Learning Techniques

Deep learning techniques for classifying images into multiple classes have made significant strides in the past few years. Nevertheless, allocating multiple classes to a single image, as seen in object detection scenarios, remains a relatively unexplored avenue of research. This study is dedicated to harnessing the potential of deep learning methods to discern various diseases within a singular leaf image of a citrus fruit. The investigation focuses on citrus leaves affected by none, one, two or all three severe diseases—anthracnose, melanose and bacterial brown spot. These leaves serve as the training dataset for ten distinct deep learning models. The performance of these models is meticulously examined, gauging their accuracy in classifying the diseases. The outcomes highlight the superiority of the DenseNet121 architecture in terms of accuracy and training duration, as it accurately identifies overlapping disease classes present in citrus leaves. Following suit, the MobileNetV2 architecture showcases comparable accuracy and a noteworthy reduction of 66% in training time.