Enhanced Sugarcane Disease Detection Using DenseNet201 and DenseNet264 with Transfer Learning and Fine-tuning

Abstract

Plant diseases have historically posed a threat to agricultural productivity and plant growth, which can have negative effects on the world's food supply. A significant commercial crop with a high net production value is sugarcane, which makes it a good candidate for disease detection. However, a variety of illnesses is causing the rate of production to decline annually. In order to diagnose diseases before they contaminate crops and save financial losses, it is essential to identify the type of infestation and the presence of the disease. This research suggests a customised DenseNet201 and DenseNet264 based sugarcane plant leaf and stem disease detection method to achieve this purpose. Images of sugarcane leaves representing both healthy and ill classes are extracted from a sugarcane dataset and used for testing. The photos are then preprocessed using techniques like resizing, contrast improvement, and grayscale conversion. Some significant characteristics are recovered from the preprocessed pictures using a customised DenseNet201 and DenseNet264 architecture with fine tuning and transfer learning. Metrics including accuracy, precision, recall, and F-measure are examined in order to assess how well the suggested method can identify diseases affecting sugarcane plants. The experimental findings demonstrate that compared to other current methods, the suggested strategy yields a greater detection accuracy of roughly 98.45%.