Hybrid Models for Plant Disease Detection using Transfer Learning Technique
Plant diseases can sometimes be so bad that there is no grain harvest at all. Consequently, there is a great need for automatic plant disease identification and diagnosis in the realm of agricultural informatics. Many approaches have been put forth to solve this problem, but deep learning is quickly...
Saved in:
Published in | 2023 10th International Conference on Computing for Sustainable Global Development (INDIACom) pp. 712 - 718 |
---|---|
Main Authors | , , |
Format | Conference Proceeding |
Language | English |
Published |
Bharati Vidyapeeth, New Delhi
15.03.2023
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | Plant diseases can sometimes be so bad that there is no grain harvest at all. Consequently, there is a great need for automatic plant disease identification and diagnosis in the realm of agricultural informatics. Many approaches have been put forth to solve this problem, but deep learning is quickly winning popularity due to its outstanding results. In this study, we assess the transfer learning capabilities of deep convolutional neural networks for the diagnosis of plant leaf diseases. We also consider how well the trained model will perform when applied to the specific task identified by our own data. One of the deep learning methods, the convolutional neural network algorithm (CNN), has been successfully applied to solve computer vision issues such as image classification, object segmentation, picture analysis, etc. In our study, we apply the Densenet201, Xception, InceptionResnetv2, NasNetMobile and VGG-16, hybrid with Support vector machine (SVM) along with a transfer learning approach to recognize diseases in images of different leaves. Instead of starting the training from scratch and arbitrarily initializing the weights, we use the pre-trained networks on the large labelled dataset, ImageNet, to build the weights. The proposed method's validity is supported by experimental findings, and it effectively detects plant diseases. The suggested method significantly outperforms other cutting-edge approaches in terms of performance like (F-scores of 0.93) for VGGG16+SVM, (F-score: 0.97) for DenseNet201+SVM, (F-score: 0.87) for Xception+SVM, (F-score: 0.97) for InceptionResnetv2+SVM, and (F-score: 0.93) NasNetMobile+SVM hybrid model on the dataset of plant images. |
---|