Detection of Banana and Its Ripeness Using Residual Neural Network

Abstract

Automatic fruit detection utilizing computer vision techniques has been carried out to help the agriculture and plantation industries. This study researches smart systems to detect bananas and ripeness classification utilizing residual neural networks. The method used to detect bananas is transfer learning from pretraned Model VGG-19. Whereas, in the bananas ripeness classification process, residual neural networks, which are trained from the start, are used. Sliding Windows is used to detect the position of bananas followed by Non-Max Suppression to summarize the results of several detected bananas. Previous studies were limited to the level of ripeness, but in this study, bananas are detected and followed by the level of bananas ripeness (raw, ripe, and overripe). This study’s data uses bananas which were mixed with other kinds of fruit. There two kinds of bananas detection architecture used in this study, VGG-19 and Restnet. After they were used to detect bananas, it was found that VGG-19 was more suitable. The results of this study are very satisfying as it is seen from the bananas detection testing percentage using VGG-19 architecture which shows 100% ripe bananas, 99 % raw bananas, and 100% overripe bananas.Keywords: Detection of banana, banana ripeness, Non-Max suppression, residual block.