Abstract:
The damage done by rice leaf diseases is detrimental to our food safety and economic
growth. Farmlands are being reduced day by day due to urbanization. Therefore it is
highly essential to have the maximum harvest within limited space. But rice leaf diseases
make it difficult to maximize the harvest. In Bangladesh brown spot, leaf scaled, rice
tungro, sheath blight, and leaf blasts are the main diseases that are commonly seen in rice
leaves. Detection of such diseases at an early stage can save a lot of crops and increase
productivity. But this cannot be achieved manually within a short time in an accurate
way. This paper comes up with a solution using CNN and transfer learning to identify the
diseases before they can spread any further. The dataset was collected and modified from
Kaggle and Mendeley Data. Our objective is to detect rice disease quickly and accurately
to aid the agricultural sector and save a lot of time and effort for the farmers.
InceptionV3, Xception, Resnet50V2, NasNetLarge and VGG16 were five transfer
learning models that we applied. CNN without transfer learning,Xception, Resnet50V2,
NasNetLarge and VGG16 accuracies are respectively 79.32%, 84%, 88%, 82%, and 81%.
Whereas the InceptionV3 model achieved better results with an accuracy of 95%.
