Deep Learning Models for Beans Crop Diseases: Classification and Visualization Techniques

Authors

  • Priyanka Sahu, Anuradha Chug, Amit Prakash Singh, Dinesh Singh, Ravinder Pal Singh

Abstract

Crop diseases highly inhibit their growth. It may cause a critical loss of yield in crops; thus, respective crop quality or quantity gets affected. This is the reason why the detection of the disease in crops plays a significant role in the field of agriculture. Detection of crop diseases using some automatic techniques is helpful as it minimizes a massive work of supervision in big fields of production. It identifies the early symptoms of diseases in crops, i.e., as when they start to become visible on the plant leaves. In this study, beans crop leaf images were used in training for the classification, with a total of 1296 leaf images. Two Deep Learning models, namely, GoogleNet and VGG16 have been used to automatically extract the features from the images fed to the trained network. For training, bean crop leaves were classified into three different categories (classes), namely, Angular Leaf Spot, Beans Rust, and Healthy. Experimental results show that GoogleNet performs better than VGG16 with an accuracy of 95.31%. Visualization approaches, namely, Visualization of Intermediate layer activations, Visualization of the CNN filter, and Visualization of Heat Maps were used for analyzing, understanding the symptoms, and localization of diseased regions in the leaves. Moreover, it helps the naïve users to understand how a convolutional neural network works internally "instead of a black box" to identify and classify the diseased regions in an image.

Downloads

Published

2021-03-01 — Updated on 2021-03-30

Versions

How to Cite

Ravinder Pal Singh, P. S. A. C. . A. P. S. D. S. . (2021). Deep Learning Models for Beans Crop Diseases: Classification and Visualization Techniques. International Journal of Modern Agriculture, 10(1), 796 - 812. Retrieved from https://modern-journals.com/index.php/ijma/article/view/670 (Original work published March 1, 2021)

Issue

Section

Articles