Non-destructive discrimination of vacuum-dried banana using image processing operation and machine learning approach

Abstract

The objective of this study was to distinguish banana samples subjected to 18 different combinations of vacuum drying and various pretreatments using innovative models built based on image textures. Banana slices were either untreated, pretreated by exposure to ultrasound or microwaves, or pretreated by immersion in 5% ascorbic acid or 5% citric acid solution. Samples were then dried in a vacuum dryer at 70 °C and one of three pressures: atmospheric, 200 mmHg of vacuum or 400 mmHg of vacuum. Samples were imaged and classified using a machine learning approach based on 1629 texture parameters in nine color channels. The dried banana samples were classified with high overall accuracies reaching 96.89% for a model including combined selected textures from images in all color channels developed using the Random Forest algorithm from the group of Trees. The confusion matrix revealed the greatest mixing of cases between banana slices dried at 70 °C + gum arabic pretreatment vs. 70 °C + microwave pretreatment and 70 °C+ 200 mmHg + gum arabic pretreatment vs. 70 °C+ 200 mmHg + microwave pretreatment. The successful results showed that the assessment of the influence of drying and pretreatment on banana slices could be carried out using image processing and machine learning in a non-destructive, objective, and effective manner.