Inspection of quince slice dehydration stages based on extractable image features

Abstract

The relation between the moisture content of the fruit and image-based characteristics was investigated. Quince samples were dried in an oven dryer at three different temperatures (40, 50, and 60°C). Several shape, texture, and colour features of the quince slices were extracted from the images. Gradual reduction was observed in all morphological features when the moisture content of the samples decreased. Regression equations between the extracted features and moisture content of the quince slices were investigated. The moisture content prediction equations based on morphological features were more precise than the textural features while colour information did not yield any satisfactory result. To exploit the morphological and textural features simultaneously, several artificial neural network models were developed to predict the drying behaviour of quince. R2 and RMSE values were determined as 0.998, 0.008%. It was concluded that the combination of the neural networks and image processing technique has the potential to determine the moisture variations.

Keywords

• machine vision
• neural networks
• moisture content
• drying
• quince