Pretrained CNN Architectures: A Detailed Analysis Using Bell Pepper Image Datasets

Abstract

In the era of artificial intelligence, automation is becoming popular in every sector. The primary sector includes the agriculture sector. Farmers are facing problems such as the identification of diseases in their plants, lack of proper treatment for the disease, climatic changes that affect their yield, and low price for their crops. In this paper, we are mainly focusing on the disease identification of bell pepper plants using deep learning architectures such as Alex Net, google net, ResNet (18,50,101), and Vgg (16,19). We also focus on the detailed study of different pre-trained CNN architectures to analyze their performance and identify which architecture is more suitable for disease classification in bell pepper. This paper also helps bell pepper farmers to identify the disease with high accuracy compared to the traditional methods of disease identification. The new automation concept helps bell pepper framers to identify diseases with less time and effort, which makes their work easier. The identification of disease at an early stage with less effort will help the farmer to increase their yield. The paper will help to understand the performance of different pre-trained convolutional neural network architectures with and without augmentation of images and also compare the performance of the architectures. Based on these comparisons, it could find out that google net is more suitable for the classification of images in bell pepper as compared to other architectures with augmentation, and vgg19 was observed to be best for the classification of images without augmentation.

Plant phenotyping relevance

ベルペッパー植物の画像から病害を分類するCNN手法を複数比較し、データ拡張の有無を含めて性能評価しているため、植物病害フェノタイピング手法が研究の中心である。

Code and data availability

The paper's phenotyping input is the public PlantVillage bell pepper (bacterial spot vs. healthy) image dataset obtained from Kaggle, used to train/evaluate the pretrained CNN architectures. No author analysis code, trained models, or supplementary deposits are mentioned.