A Multi-Crop Disease Detection and Classification Approach using CNN

Being an agricultural country, the well-being of plants plays a vital role in the agricultural yield. Various diseases and disorders affect the quality and quantity of the production, thus the intelligent methods for disease detection in crops is the need of hour. We have proposed a robust and generalized approach that detects diseases in multiple crops by utilizing the baselines of existing CNN models. We have proposed the variants of ResNet-152 and Inception-v3 for detection of diseases in essential crops like rice and corn. We have used publicly available datasets having three different diseases in rice and both infected and healthy leaves of corn. The proposed method has achieved accuracy of 97.81% and 97.48% by employing variants of InceptionV3 and ResNet152 respectively for corn crop. To understand the diversity of diseases, we categorize the rice disease images into major and minor subsets.The proposed ResNet152 variant has achieved accuracy of 99.10% and 82.20% for the major and minor disease subsets respectively. Experimental results of proposed approach indicate robustness in disease detection for multiple crops.