A Novel Model Proposal and Comparative Analysis of Deep Learning Techniques for Classifying Cercospora beticola and Erysiphe betae Diseases on Sugar Beet Leaves

This study introduces a novel approach utilizing a convolutional neural network (CNN) architecture for the detection and classification of Cercospora beticola and Erysiphe betae diseases, aiming to enhance both the quantity and quality of sugar beet yield, a pivotal commodity in agriculture. The research focuses on disease identification and plant categorization, leveraging deep learning (DL) techniques for sustainable agricultural practices. Delayed detection and treatment of these diseases pose significant threats to harvest productivity, emphasizing the importance of timely intervention. Timely and accurate disease detection is crucial for improving sugar beet yield and quality for agricultural production. This study employed DL methods to classify sugar beet leaf images into healthy or diseased categories, followed by sub-classification into Cercospora beticola or Erysiphe betae. The proposed model's efficacy was evaluated through comparative analysis with established models such as the Visual Geometry Group networks (VGG16 and VGG19), InceptionV3, AlexNet, and ResNet50, renowned for their robust performance in image classification tasks. The dataset consisted of 4128 samples covering healthy and diseased sugar beet leaves, further classified as Cercospora beticola and Erysiphe betae. Additionally, the performance of the proposed model was compared with other models in terms of train time. Remarkably, although transfer learning is not implemented in the proposed model, it achieves 98% accuracy, 96% precision, 100% recall, and 98% F1-score, exceeding transfer learning models. This study advocates adopting a CNN model with a light-weight structure, facilitates rapid assembly, and has high recognition sensitivity of disease classification.