Journal of Information Technology & Software Engineering
Open Access

Abstract

Capillary Dynamolysis Image Discrimination Using Neural Networks

Mehmet S. Unluturk, Sevcan Unluturk, Fikret Pazir and Firoozeh Abdollahi

Quality differences between organic and conventional fresh tomatoes (unprocessed) and frozen tomatoes (processed) are evaluated by using a capillary rising picture method (capillary dynamolysis). The best pictures showing the differences most sharply between organic and conventional samples were prepared with 0.25-0.75% silver nitrate, 0.25-0.75% iron sulphate and 30-100% sample concentration. But visual description and analysis of these images is a major challenge. Therefore, a novel methodology called Gram-Charlier Neural Network methodology (GCNN) has been studied to classify these images. Two separate GCNNs have been created for fresh and frozen cases. They are trained with the pictures of organic and conventional tomato samples from these two cases. The 2048 x 1536 pixel chromatogram images were acquired in a lab and cropped to 1400 x 900 pixel images depicting either a conventional tomato or an organic tomato for each case. A set of 20 images from each case was utilized to train each Gram- Charlier Neural Network. A new set of 4 images from each case was then prepared to test each GCNN performance. In addition, Hinton diagrams were utilized to display the optimality of the GCNN weights. Overall, the GCNN achieved an average recognition performance of 100%. This high level of recognition suggests that the GCNN is a promising method for the discrimination of capillary dynamolysis images and its performance does not depend on whether the tomato sample is fresh or frozen.