Abstract: In India the major percentage, around 70% of the population relies on agriculture. In order to produce a healthy yield, the identification of the plant diseases is important. The diseases that affect plants cannot be manually observed by the farmers. It's a troublesome and a very time-consuming process for them to detect the plant diseases manually. Hence, image processing and machine learning models can be used to detect the plant diseases at an early stage. Plant disease detection majorly relies on the image classification concept of machine learning. The benefit of using transfer learning is the model being used has a pre-build knowledge on which the model starts from different patterns. These patterns been learnt while training on a different dataset used for a different problem which is similar to the problem being solved using the same model. Automatic recognition and classification of various diseases of a specific crop are necessary for accurate identification. This study mainly concentrates on the transfer learning phenomenon based on four different pre-trained models such as VGG-16, ResNet-34, ResNet-50, and ResNet-50 v2 and then compared the four models based on various standard evaluation metrics such as accuracy, recall and precision. The dataset considered for the study includes the various diseased and healthy leaves of different plants and crops such as Apple, Corn, Pepper, Potato, and Tomato, etc.
The goal is to study and recognize the best and most efficient method that can be used to detect the diseases in the crops. Various methodologies with their corresponding accuracies along with a detailed comparison, discussion of the features, input parameters and experimental setup will be discussed. The comparative study of the pre-trained models in terms of their performance will be carried out.

Keywords: Deep learning, Transfer learning, Pre-trained models, Image classification

| DOI: 10.17148/IARJSET.2022.9637