Premium
Artificial Neural Network‐Based Image Analysis for Evaluation of Quality Attributes of Agricultural Produce
Author(s) -
Rafiq Aasima,
Makroo Hilal A.,
Hazarika Manuj K.
Publication year - 2016
Publication title -
journal of food processing and preservation
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.511
H-Index - 48
eISSN - 1745-4549
pISSN - 0145-8892
DOI - 10.1111/jfpp.12681
Subject(s) - rgb color model , artificial neural network , artificial intelligence , projection (relational algebra) , computer science , pattern recognition (psychology) , image quality , mean squared error , image processing , identification (biology) , computer vision , image (mathematics) , mathematics , statistics , algorithm , biology , botany
The present study aimed to apply artificial neural networking for quantification of quality attributes of agricultural commodity based on color and size. Three feed forward neural network models ( NN ) were developed: first for conversion of RGB to L *, a * and b * values ( NN 1 ), second for identification of ripening stages of tomato and third for correlating projection area/size of tomato with weight. Results showed that NN 1 was able to convert RGB to L *, a * and b * values with accuracy percentage of 99%. The best results showed with excellent abilities were at 30 hidden units with R 2 of 0.9980 and mean squared error (MSE) of 0.00021, whereas NN 2 classified tomatoes in three ripening stages with an accuracy of 96%, with 30 hidden neurons and a 100% classification was performed when a threshold value of 0.7 was used. In addition, NN 3 was able to correlate area with weight with an accuracy of 99%, with three hidden neurons. Practical Applications Computer vision system‐based measurement of color and projection areas has been considered for the assessment of tomato quality. Image processing, a nondestructive method for evaluation of quality attributes of agricultural produce, serves as a best tool for measurement of size and color of nonuniform products. The color of whole surface can be evaluated using image analysis. Image processing was used as the basic tool and their functional relationships are expressed in terms of artificial neural network model.