Paddy Leaf Disease Detection using Thermal Images and Convolutional Neural Networks

India is an agriculture-based economy with agriculture being the primary sector of economy and contributing to more than 18% to total GDP and employing approximately 60% of total population of the country directly and indirectly. Also, India is one of the leading producers of food crops like wheat, rice, millets and pulses with wheat and rice being most cultivated ones making it second largest producer of wheat and rice just behind China. Till date, in many parts of the country and world, rice is staple food for millions of people. Yet, due to various reasons, agricultural sector is lagging far behind other sectors in terms of embracing the latest technological advancements. This results in lowered production and productivity which directly impacts the agricultural output. Crop disease and crop failure being one of the major reasons for less agricultural output. There are more than 35 diseases associated with paddy crops but 8-10 being most commonly encountered. The main aim of this paper is to provide best detection results to these diseases and also to try and automate the whole process of disease classification and detection using Convolutional Neural networks (CNN) and thermal imagery techniques. Analysis and comparison of results has been done on various algorithms- Generalized CNN model, CNN with optimizations using SVM as final layer and transfer learning models like VGG19, ResNet 152V2, Inception V3, MobileNet V2.