Enhancing Disease Classification in Paddy Fields: A Stratified Cross-Validation Approach

Abstract

Abstract In India rice (Oryza sativa) is an important food staple. It is the third highest production among worldwide. Rice is highly rich in calorie so morethan one-fifth of the population in world consumed Rice. Abiotic and Biotic factors plays a vital role in production of Rice as it affects and causes more damage. Biotic factors like diseases and pests leads to 70% of loss in crop production. Identifying diseases in early stage is a tedious concern for every farmer. Once the disease is predicted in early stage, solution or necessary steps can be taken to reduce the damage. Agricultural officers or External experts have to check manually and give the remedial solutions for this issue. But, due to lack of available resource external experts cannot visit field for every time. So identifying the correct disease is very difficult. One solution for this concern is by using latest advancement in technology. Convolutional Neural network is mainly used for classifying images. We have taken 10,407 labelled images for training the model and 3,469 images for testing the model. We used transfer learning model namely InceptionV3, ResNet50, VGG16, MobileNetV2, and EfficientNetB0 to obtain the result. Finally CNN model ResNet is applied with Stratified Cross-validation fastai techniques. K-fold cross validation strategy obtained an highest accuracy of 98.81% which is more accurate than traditional Transfer learning models.