Fruit identification employing computer vision based on artificial neural network backpropagation algorithms

  • I.A. Kamil
  • A Fagbadegun

Abstract

Fruit identification is an essential process in automatic fruit sorting and grading. Automatic fruit recognition usually employs image processing and computer vision techniques. Various soft computing techniques have recently been deployed by researchers in the learning module of object recognition algorithms in which Artificial Neural Network (ANN) Backpropagation(BP) algorithms are the mostly adopted for fruit recognition. Performance of different ANN Backpropagation algorithms vary for different tasks. The aim of this research is to determine the most appropriate ANN BP algorithm for fruit recognition.

130 samples of fruits comprising 65 sweet orange and 65 coconuts were collected. The images of the collected fruits were captured using LW-IC500 light wave webcam camera. Three main features (shape, colour and texture) of the captured images were extracted and used as inputs to train thirteen different ANN BP algorithms in a supervised manner to identify the type of fruit. Simulations were carried out in Matlab environment and the performances of these algorithms in terms of the number of epochs, accuracy of recognition, Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) were measured and compared.

The results showed that One Step Secant algorithm produced a recognition accuracy of 93.6% with a validation of 0.26136 at 3 epochs while Gradient Descent with Adaptive Learning BP algorithm gave a recognition accuracy of 71.7% with a validation of 0.11629 at 74 epochs. Their RMSE were 0.2627 and 0.3250 respectively while the MAE were 0.1283 and 0.2701 respectively.

It has been shown that One Step Secant BP outperformed all other BP algorithms in fruit recognition system. Its lowest number of iterations leads to fast operation and therefore supports real-time application.

Keywords: Artificial Neural Networks, Backpropagation Algorithms, Fruit Recognition, Feature Extraction

Published
2019-03-20
Section
Articles

Journal Identifiers


eISSN: 1596-3233