Maize Leaf Disease Identification Using Deep Learning Techniques: A Case Study in Bangladesh

Abstract

This paper presents research on the deep learning methods and how the methods can be used in detecting three of the common corns on the leaves. A dataset was collected containing more than 4,000 actual pictures of contaminated and wholesome curvy leavesof corn. The CNN, MobileNetV2, VGG16, VGG19 and InceptionV3 deep learning models were trained and tested on this dataset. The CNN model attained an accuracy of 83%, followed by MobileNetV2 97%, VGG16 87%, VGG19 85% and InceptionV3 96%. These results indicate that CNN-based models can be an effective and reliable method to identify diseases that potentially are threatening maize farmers in Bangladesh and promote agricultural resilience at a scale and cost-effective approach. This study shows that modern technology has the ability to redefine agriculture. With the provision of more accurate, fast and cost-effective disease detection, such models of mobile deep learning, as MobileNetV2and InceptionV3, provide practical tools that could be used by farmers in the field. In terms of mobile or edge deployment, MobileNetV2 in particular will provide a feasible means of real-time disease detection that can be applied using smartphones by farmers. It was converted to TensorFlow Lite format and then successfully deployed in a mobile application.