Abstract:
Tea plants have a lengthy lifespan, which makes disease management challenges. Tea
leaves are commonly afflicted by bacterial, viral, or fungal infections, which significantly
decrease tea production. Identifying disease tea leaves is critical to sustaining worldwide
tea demand for a vast population. Early disease detection can assist in reducing losses.
However, disease tea leaf detection was limited to image backgrounds and image-capture
conditions. In the field of disease tea leaf identification, the convolutional neural network
(CNN)-based model is the most popular topic of research. However, existing some CNNbased models have low recognition rates on independent datasets and are limited to
learning large-scale network parameters. We proposed a novel CNN-based model to
identify disease tea leaf in this study by decreasing network parameters. A number of CNNbased models, namely InceptionV3, MobileNetV2, MobileNet, Vgg16, Xception, and
InseptionResNetV2, are trained to distinguish between healthy tea leaves and diseased tea
leaves using a new dataset of 2000 tea leaf images. Out of a total of 100% data, we used
75% of image data for training and 25% of image data for testing. By using MobileNet we
have achieved the highest accuracy which is 98%. To increase the image's overall accuracy,
we used the data augmentation approach. In comparison to other approaches, the
combination of transfer learning and data augmentation produces the highest efficiency.