No Detail Found!

Empirical study on fungal diseases in various plants using different deep learning approaches

Author(s)
G.SRAVANTHI Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, AP, India.
P.SAI KIRAN Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, AP, India, Department of Computer Science and Engineering Prasad V Potluri Siddhartha Institute of technology ,AP, India.
Abstract
The productivity of the agriculture decreasing day by day due to various factors. The quality of the agriculture is impacted by the pests and diseases infected on the plants. Many researches proposed traditional image processing techniques to identify the diseases and design a recommendation system but all these are expensive and inaccurate systems, which are not affordable by farmers. When a new era known as “Machine Learning” as evolved, researchers extracted necessary features from the plant images and converted them into csv files to classify the diseases in plants using these approaches. Even these approaches are time consuming, so researchers are moved to deep learning approaches in which the intermediate steps takes places automatically in between input and output like pre-processing feature extraction. In this paper, the model studied about different deep learning approaches available on the disease detection system in various plants. Among the deep learning models, the researchers who have implemented transfer learning approaches have obtained high accuracy and few have achieved good kappa cohen values also. In traditional models the system needs to assume random weights for the neurons to produce the dot vectors of every layers. These assumed values some times may give more error rate which results in back propagation. Transfer learning helps the model to get the optimal weights without assumptions of each neuron from the pre-trained model it implements. The major advantage of any pretrained model lies in faster training of network with millions of different categories images. One of the popular dataset used by most of the researchers for this study is “ImageNet”. Using the concept of transfer using the ImageNet irrespective of any pre-trained model implemented the accuracy lies in between 92% to 96%.
Publication Details
Page(s) 6640-6651
DOI DOI not available
Published Journal: Journal of Theoretical and Applied Information Technology, Volume: 100, Issue: 21, Year: 2022

Request the full text from PASTIC

Keywords
Transfer Learning, segmentation, Image Enhancement, Annotation, Augmented Data
0

Citations

0

Downloads

95

Views