VGG16-Based MaizeLeafNet Model for an Efficient Multiclass Classification of Maize Leaf Diseases
Archana Saini, Kalpna Guleria, Shagun Sharma
Abstract
Maize holds substantial importance as a cereal crop in India, serving both as a staple food for human consumption and as fodder for livestock. The presence of diseases that affect maize leaves poses a significant threat, capable of causing substantial reductions in crop yield and quality. The timely detection and effective management of these diseases are imperative for ensuring the preservation of agricultural productivity. India is a country with a sizable and expanding population, and a major portion of its food comes from maize. A steady and safe supply of food for the populace is ensured by keeping maize crops free from diseases. Crops that are resistant to disease are crucial for providing for the dietary needs of an expanding population. Hence, in the proposed work, a deep learning-based VGG16 model called MaizeLeafNet has been proposed, for the early prediction of the maize leaf diseases. It is well known for having powerful feature-learning capabilities. The collection of a large dataset featuring a range of images of diseased maize leaves ensures a representation of the challenges faced in real-world farming scenarios. Measures like accuracy and loss are used to carefully evaluate the model's performance in multiclass classification. Through the use of visualization techniques, the model's interpretability is enhanced, improving understanding of disease patterns and providing new perspectives on the features that have been learned. Developing crop disease classification frameworks that use state-of-the-art neural network designs, such as VGG16 is a significant step towards creating resilient and sustainable farming methods. This article compares the loss and accuracy at different epochs in the training and validation stages. At the epoch value of 10, the accuracy of 90.34% and the loss of 0.4226 have been identified during validation, whereas the accuracy of 98.79% and the loss of 0.0380 resulted during training. These outcomes show that as the epoch increases to 20, the training accuracy and testing accuracy has been increased to 99.68% and 91.55%, respectively, whereas, the training and validation loss also decreased to 0.0039 and 0.3727, respectively.