Detection and Classification on Plant Disease using Deep Learning Techniques

Abstract

Plant diseases are a major problem for the agriculture industry because they can cause large crop losses and jeopardize food security. Deep learning approaches have demonstrated encouraging results in automating plant disease diagnosis and detection in recent years. In the context of plant disease diagnosis, this study examines the efficacy of two well-known convolutional neural network architectures: DenseNet121 and VGG16. Plant Village datasets are used for pretrained and fine-tuning of the DenseNet121 and VGG16 architectures, respectively. The dataset includes Images of both healthy and sick plants. To guarantee the models' resilience and generalizability, the dataset include 15 different classes and 4 types of plants namely Tomato, Potato and Pepper Bell. We compare the accuracy, precision, recall, and F1-score of DenseNet121 and VGG16 for plant disease classification through extensive testing and analysis. To determine if they are practically feasible for use in real-world applications, we also examine their model complexity and computing efficiency. Our findings show that DenseNet121 and VGG16 can both correctly diagnose plant diseases in a variety of species. Although DenseNet121 outperforms VGG16 in terms of overall accuracy and computational efficiency, both models obtain high accuracy rates. Additionally, DenseNet121 has superior generalization performance, especially in identifying uncommon or underrepresented illness classes. All things considered, this work emphasizes the promise of deep learning models-DenseNet121 in particular-as useful instruments for automated plant disease identification and points to directions for further investigation to improve the efficiency and scalability of such systems for real-world use in agriculture