Deep Learning Techniques for Accurate Classification of Rice Diseases: A Comprehensive Study

Agriculture is a paramount source of revenue and sustenance worldwide. In India, for instance, 58% of the population lives off agriculture. On top of that, a range of 20% to 40% of global harvests are lost yearly to pests and illnesses. Crop safety relies on early and precise plant disease identification. Early plant disease detection may save yields and reduce starvation and expenditure. In India's regions, rice is farmed as a staple. Diseases have a severe damaging influence on rice harvests, resulting in considerable losses for the agricultural business. Typically, identification is performed by visual inspection or laboratory analysis. While the lab test is time-consuming and may not yield rapid results, visual observation requires expertise and is prone to viewers’ subjectivity. Accurate and reliable methods to diagnose rice plant sickness are highly wanted by plant pathologists. In this regard, the visual domain is critical to implementing automatic identification systems to catch plant diseases timely using machine learning techniques. In this work, we propose a reliable convolution neural network-based approach for identifying illnesses in rice plants. Two well-known rice diseases caused by fungus, such as leaf blasts and leaf brown, have been investigated. Therefore, the data is separated into blast, brown, and healthy. The blast category comprises 2988 photos, with Brown and Healthy counting 2988 and 1532 samples. This work offers a comprehensive assessment of CNN-based methods relying on transfer learning paradigm over rice plant disease classification tasks.