Banana Ripeness Classification via Thermal Imaging with Transfer Learning

Banana fruits are graded manually as unripe, ripe and overripe. This research work has been presented to predict the ripening level of banana fruit using digital thermal imaging and deep learning algorithm. Thermal image analysis on banana were carried over to predict its ripening stage by the transfer learning approach of existing deep learning models. In total, 850 thermal images of banana from 97 different fruits were used for the experimentation, and they are classified into three maturity levels. The experimental analysis demonstrates that the outcomes exhibit training accuracy of 96.46%. This is then used to obtain a confusion matrix which shows an accuracy of 97.7% and have useful applications for small scale vendors, large scale vendors and export industry. Further the study can be extended to improve the accuracy by using more advanced or customized CNN models.