A Comparative Analysis on Fruit Freshness Classification

Abstract

Automatic classification of food freshness plays a significant role in the food industry. Food spoilage detection from production to consumption stages needs to be performed minutely. Traditional methods which detect the spoilage of food are slow, laborious, subjective and time consuming. As a result, fast and accurate automatic methods need to be introduced to industrial applications. This study comparatively analyses an image dataset containing samples of three types of fruits to distinguish fresh samples from those of rotten. The proposed vision based framework utilizes histograms, gray level co-occurrence matrices, bag of features and convolutional neural networks for feature extraction. The classification process is carried out through wellknown support vector machines based classifiers. After testing several experimental scenarios including binary and multi-class classification problems, it turns out to be the highest success rates are obtained consistently with the adoption of the convolutional neural networks based features.