A Novel Zero-Shot Learning Approach for Multi-Classification of Orange Sooty Mold Disease Severity Levels
Manni Kumar, Nitasha, Manpreet Kaur, Monika Monika, Nivedita Sharma
Abstract
Orange sooty mold disease is a common problem in the citrus industry that affects the quality and yield of oranges. The current study proposes a zero-shot learning approach for the multi-classification of orange sooty mold disease severity levels using a dataset of 20,000 images collected from various orange fields. The proposed approach utilizes a pre-trained CNN model to extract features from the images, which are then mapped to a semantic space constructed using the WordNet hierarchy to represent the relationships between different severity levels of the disease. The classification is performed using a linear classifier trained on the semantic space, which enables the model to classify images into severity levels that were not seen during training. The proposed approach achieved an overall accuracy of 95.88% in the multi-classification of the six severity levels of the disease. The findings show that the technique that was presented for the multi-classification of orange sooty mold disease severity levels is successful. The proposed approach provides a valuable tool for growers and researchers to identify and monitor the progression of the disease in orange crops. The approach has the potential to revolutionize the way that orange sooty mold disease is monitored and managed in the citrus industry, providing a more efficient and accurate method for disease classification and monitoring. In a further study, one possibility for further enhancing the effectiveness of the suggested method is to investigate the employment of more sophisticated methods.