Wheat Fungal Disease Detection Using a U-NET Deep Learning Model

Abstract

world’s important cereal it often falls victim to diseases like Fusarium Head Blight (FHB) and rust. These diseases can seriously reduce wheat yields and threaten food supplies. Traditionally, identifying these diseases relies on experts carefully examining the plants by hand can be quite subjective. As a result, early and precise diagnosis is still difficult, hence the impetus for automated solutions. We suggest an automatic deep-learning solution for wheat disease segmentation and detection to overcome these issues. We utilize Convolutional Neural Networks (CNNs) to obtain stable image-based classification and accurate symptom localization of the disease. In particular, EfficientNetB0 is used for disease classification because it is efficient and accurate, while a specially designed U-Net model is utilized for pixel-wise segmentation of infected areas. The experimental environment uses Kaggle's free GPU power for model training, OpenCV for image pre-processing, and TensorFlow/Keras for implementation. Significant hyperparameter optimization (tuning learning rates and batch sizes) is done to achieve optimal performance. In a series of experiments on a wheat disease image dataset, the suggested system.