Rapid Detection of Changes in Image Textures of Carrots Caused by Freeze-Drying using Image Processing Techniques and Machine Learning Algorithms

Abstract

The objective of this study was to evaluate the differences in texture parameters between freeze-dried and fresh carrot slices using image processing and artificial intelligence. Images of fresh and freeze-dried carrot slices were acquired using a digital camera. Texture parameters were extracted from slice images converted to individual color channels L, a, b, R, G, B, X, Y, and Z. A total of 1629 texture parameters, 181 for each of these color channels, were obtained. Models for the classification of freeze-dried and fresh carrot slices were created using various machine learning algorithms, based on attributes selected from a combined set of textures extracted from images in all color channels (L, a, b, R, G, B, X, Y, and Z). Using three different feature selection methods (Genetic Search, Ranker, and Best First), the 20 most effective texture parameters were determined for each method. The models with the highest classification accuracy obtained by applying various machine learning algorithms from Trees, Rules, Meta, Lazy, and Functions groups were determined. The classification successes obtained with the parameters selected from all three different feature selection algorithms were compared. Random Forest, Multi-class Classifier, Logistic and SMO machine learning algorithms achieved 100% accuracy in the classification performed with texture features obtained by each feature selection algorithm.