MLTSDC: Multi Level Transformer Based Sugarcane Disease Classifier

Abstract

Sugarcane is a vital crop that is not only a primary source of the sugar but also a major contributor of bio fuel. However, unfortunately, due to various diseases, huge agricultural yields of sugarcane are lost every year. This amount can be easily saved if a farmer can identify the diseases in their early stage. For this purpose, numerous models have been proposed in recent years. However, these models cannot produce acceptable impacts in real-world applications due to background noise, environmental factors, and resource constraints. Therefore, this paper proposes a Multi Level Transformer Based Sugarcane Disease Classifier (MLTSDC) model to solve this problem. The proposed MLTSDC model utilizes two levels of classification. The first level identifies the presence or absence of abnormal features, while the second maps them to their diseases. This enables the model to learn and identify the key features of diseases. Moreover, the MLTSDC model also incorporates the transformer’s self-attention mechanism to minimize the effect of background noise. Experiments conducted on publicly available datasets reveal that the proposed MLTSDC model works better than other existing models. The proposed model achieves the highest classification accuracy 98.8% for different diseases affecting real-world sugarcane leaves.