Performance Analysis Disease Detection Using Image Processing for Feature Extraction of Machine Learning Algorithms on Rice Leaf
Abstract
Rice is the staple food for a large part of the world‘s population with such diseases interfering with the production of the staple food like bacterial blight, brown spot and smut disease among others. Identification of these diseases at an early stage and correct diagnosis can easily reduce the loss in yield and quality. Conventional approaches used in the identification of diseases in rice crops are executed through manual assessment which takes considerable time to complete hence compromising the identification of the diseases early enough to enable treatment prior to fruition of the disease‘s aunts. This has highlighted the need to implement an automated, efficient, and accurate means of diagnosing rice leaf diseases at their early stages.
The major goal of this project was the creation of a DL model that effectively identifies images of rice leaf diseases allowing for fast and efficient disease diagnostics and control. CNN was applied on a dataset of images of rice leaf diseases, each labeled in the three categories. The approach included data preparation (resizing the image, normalizing, augmenting), training the model using various architectures (Normal CNN, CNN with Data Augmentation, CNN with Hyperparameters tuning), and assessing the model using
correct and incorrect metrics. The results performing Normal CNN model pegged an accuracy level of 90% higher than CNN models with data augmentation and hyperparameter tuning levels of 85% and 75% respectively.
In conclusion, based on the evaluation of all models on the validation set, it was concluded that the Normal CNN model is the most suitable for this application in terms of its accuracy and the ability to generalize to new images that have not been used during the training process. This paper demonstrates how machine learning can be applied in the process of disease management in agriculture. The developed model not only improves the diagnosis speed and accuracy but also provides a new direction to diagnose other diseases in crops, creating great prospects for the development of agricultural practices and food security around the world
