3A Comparative Study of Pre-trained Transfer Learning Models in Convolutional Neural Networks for the Prediction of Diseases in Plant Leaves

Abstract

Pest infestation is the biggest problem crops confront, reducing yield and food quality. Automatic plant disease detection technology is crucial to agricultural operations because it encourages productive growth and increased yields. Neural networks (NN) are frequently employed in learning approaches to manage imaging applications. The study is to evaluate and recommend a prediction model for leaf diseases employing Convolutional Neural Networks(CNN) pre-trained Transfer Learning Models. The pre-trained transfer learning models in Deep Learning (DL) are utilized in this study to predict the leaf diseases in plants. CNN's pre-trained models, including ResNet50, VGG16 and EfficientNet are used to categorize and predict damaged plant leaves. Jupyter Notebook, a Python-based software environment, is utilized for implementation. The Plant Village dataset was created with the intention of providing effective methods for the identification of 39 distinct plant diseases. It contains 61,486 images of landscapes and foliage from plants. This collection includes plant ailments caused by bacteria, fungus, and other causes.The model's experimental outcomes demonstrate that the efficiency of CNN models and the results findings shows that the EfficientNet outperforms the ResNet50 and VGG16 in predicting the plant leaf diseases. The study explores and examines the effectiveness of using several sophisticated CNN’s to improve the accuracy of leaf disease detection.