EARLY DETECTION OF PLANT DISEASES USING A HYBRID ENSEMBLE FEATURE SELECTION WITH DEEP NEURAL NETWORK FOR MODERN AGRICULTURE

Authors

  • Rajendra Prasad Bellapu , Ramashri Tiruamala , Rama Naidu Kurukundu

Abstract

Plants play a significant role in Indian agriculture as well as the economy of the country. However, the expected
growth of plants are affected by diseases, which may cause complete damage to leaf, fruits, flower, and stem,
which also leads to economic losses in agriculture. Therefore, plant disease detection is an essential task for
improving crop quality and production process. Researchers developed popular techniques, namely Support
Vector Machine (SVM) and Convolution Neural Network (CNN), to recognize plant diseases. However, the
classification accuracy is diminished due to the high curse of dimensionality with redundant data. Feature
selection techniques are developed to address these issues, but single feature selection techniques, namely
ReliefF, F-score are unstable in nature, which affects the classification accuracy for various subsets of features.
So, as to settle all these issues, a hybrid ensemble feature selection technique is introduced in this research
study. The input images are pre-processed using a multi-scale retinex algorithm, where the segmentation of leaf
images is carried out by using Kernel Fuzzy C Means (KFCM), and affected area segmentation is carried out by
using the multilevel Otsu Thresholding technique. The features are extracted using a hybrid feature extraction
technique, and optimal features are selected using the ensemble feature selection technique with Mutual
Information (EFS-MI). Finally, Deep Neural Network (DNN) is developed to categorize the healthy and
affected leaves of Plant Village Dataset (apple and potato) and collected dataset (rice and groundnut). The
experimental results proved that the proposed DNN achieved 98.77% of accuracy while existing multi-class
SVM (M-SVM) achieved 97.03% of accuracy on potato data.

Downloads

Published

2020-12-01

How to Cite

Rajendra Prasad Bellapu , Ramashri Tiruamala , Rama Naidu Kurukundu. (2020). EARLY DETECTION OF PLANT DISEASES USING A HYBRID ENSEMBLE FEATURE SELECTION WITH DEEP NEURAL NETWORK FOR MODERN AGRICULTURE. International Journal of Modern Agriculture, 9(4), 428 - 448. Retrieved from https://modern-journals.com/index.php/ijma/article/view/232

Issue

Section

Articles