Ripeness Classification of a Bunch of Bananas Using a CNN

Bananas are one of the richest sources of vitamins and minerals in the world and bananas are also one of Thailand's most economically significant fruit crops. Images of a bunch of bananas were used to train a Convolution Neural Network (CNN). The convolutional layers applied different filters as feature detectors to create a feature map representing the detected input features. Then the features were shaped into one-dimensional features and sent to an Artificial Neural Network (ANN) for the banana ripeness classification. The results showed that the classification was quite accurate. Using the same set of images for both training and testing, the CNN achieved the accuracy of 100.00%. The average classification accuracy of 91.21% was achieved when testing with images excluded from training. The classification accuracy from class 0 to class 4 was 93.94%, 90.00%, 87.50%, 89.47% and 92.31%, respectively.