Open Access

ARTICLE

Dates Fruit Recognition: From Classical Fusion to Deep Learning

by Khaled Marji Alresheedi¹, Suliman Aladhadh², Rehan Ullah Khan², Ali Mustafa Qamar^1,3,*

1 Department of Computer Science, College of Computer, Qassim University, Buraydah, Saudi Arabia
2 Department of Information Technology, College of Computer, Qassim University, Buraydah, Saudi Arabia
3 Department of Computing, School of Electrical Engineering and Computer Science (SEECS), National University of Sciences and Technology (NUST), Islamabad, Pakistan

* Corresponding Author: Ali Mustafa Qamar. Email:

Computer Systems Science and Engineering 2022, 40(1), 151-166. https://doi.org/10.32604/csse.2022.017931

Received 17 February 2021; Accepted 18 April 2021; Issue published 26 August 2021

Abstract

There are over 200 different varieties of dates fruit in the world. Interestingly, every single type has some very specific features that differ from the others. In recent years, sorting, separating, and arranging in automated industries, in fruits businesses, and more specifically in dates businesses have inspired many research dimensions. In this regard, this paper focuses on the detection and recognition of dates using computer vision and machine learning. Our experimental setup is based on the classical machine learning approach and the deep learning approach for nine classes of dates fruit. Classical machine learning includes the Bayesian network, Support Vector Machine, Random Forest, and Multi-Layer Perceptron (MLP), while the Convolutional Neural Network is used for the deep learning set. The feature set includes Color Layout features, Fuzzy Color and Texture Histogram, Gabor filtering, and the Pyramid Histogram of the Oriented Gradients. The fusion of various features is also extensively explored in this paper. The MLP achieves the highest detection performance with an F-measure of 0.938. Moreover, deep learning shows better accuracy than the classical machine learning algorithms. In fact, deep learning got 2% more accurate results as compared to the MLP and the Random forest. We also show that classical machine learning could give increased classification performance which could get close to that provided by deep learning through the use of optimized tuning and a good feature set.

---

Citations