Comparison of Support Vector Machines and ShuffleNet for Detection of Rice Species


Abstract views: 34 / PDF downloads: 38

Authors

  • Oznur Ozaltin Ataturk University

DOI:

https://doi.org/10.59287/icriret.1378

Keywords:

Deep Learning, Classification, Convolutional Neural Networks, Hyperparameters, Machine Learning

Abstract

The objective of the current investigation is to identify the optimal approach for the automated identification of rice species based on images. Two types of datasets have been utilized. The initial dataset comprises a total of 75,000 images, including five different rice species, namely Arborio, Basmati, Ipsala, Jasmine, and Karacadag. The dataset is a balanced set comprising 15,000 images for each type. During this dataset's analysis phase, ShuffleNet, one of the pre-trained architectures in Convolutional Neural Networks (CNNs), was used. The second dataset was created by extracting features from the same rice images. 106 distinct features were acquired, which include morphological, shape, and color features as the main features. The previously extracted features have been analyzed for the classification of rice species using Support Vector Machines (SVM), which have a quadratic kernel function. Moreover, the two datasets have been obtained from the URL https://www.muratkoklu.com/datasets/. Additionally, 5- fold cross-validation has been applied for both ShuffleNet and SVM to avoid overfitting. Based on the empirical findings, ShuffleNet has achieved an accuracy rate of 99.8%, while SVM has acquired an accuracy rate of 99.9%. While the results exhibit minimal differences, the optimal algorithm choice may be contingent upon the researcher's level of proficiency in the field.

Author Biography

Oznur Ozaltin, Ataturk University

Department of Mathematics, Faculty of Science, Erzurum, 25240, Turkey

Downloads

Published

2023-08-29

How to Cite

Ozaltin, O. (2023). Comparison of Support Vector Machines and ShuffleNet for Detection of Rice Species. International Conference on Recent and Innovative Results in Engineering and Technology, 92–97. https://doi.org/10.59287/icriret.1378