Learning

—A crucial part of the crop protection system in the early and accurate identification of healthy and unhealthy plants. The orthodox methods of identification involve either visual inspection or laboratory testing. Visual inspection involves experience and can vary depending on the individual, which could lead to an error, but laboratory testing takes time and might not be able to give the results quickly. Therefore, in this paper, we propose an image-based machine learning technique to recognize and classify healthy and unhealthy plants. In this work, we have focused solely on the rice plant (Oryza Sativa). The original dataset is available on Kaggle, which includes images of both healthy and unhealthy rice plants. Dataset consists of 501 healthy rice plants and 505 unhealthy rice plants. After validation, we obtained a total of 900 images, including both healthy and unhealthy rice plants. There are 4 models that we use in this experiment: VGG16, VGG19, ResNet50, and InceptionV3. In this project, we tried data augmentation and regularization to improve the performance of our program. After regularization, the results that were obtained improved. However, the results we got when we included data augmentation were worse, so we opted to solely apply regularization. The model that provides the best accuracy for the loss model is VGG19 with 84.4% accuracy and 55.1% loss. The early identification of healthy and unhealthy rice plants using this model could serve as a preventative measure as well as an early warning system. It might also be expanded to create a model for identifying rice plants' health in the actual agricultural fields.