KLASIFIKASI TINGKAT KEMATANGAN BUAH PISANG DENGAN ALGORITMA CNN

Abstract

Bananas are the most widely produced fruit in Indonesia, which is around 9.69 million tons in 2024. With this amount of data, bananas have quite promising economic value. The production of large quantities of bananas according to the previous data needs to be carried out a rapid distribution process, because the time the bananas after harvesting may only last about 5 to 7 days in normal temperatures before the fruit rotting process will finally occur. For this reason, it is necessary to carry out a quick banana classification process, the process is carried out automatically using a machine. This study elaborates on the capabilities of the CNN algorithm  in the classification of bananas. Dataset was taken from Keagle's open source as many as 3000 image data. In their classification, researchers divided bananas into three levels of fruit ripeness, namely raw, ripe, and overripe. The study used a CNN model  consisting of several layers consisting of a 2D convolutional layer (Conv2D), a 2D pooling layer (MaxPooling2D), a flatten layer, a fully connected layer (Dense), and a Dropout layer. After the model creation process is complete, the model will be tested for accuracy with  the Confusion matrix method. In the 3rd experiment the model produced the highest level of accuracy in its trials, with 300 test images resulting in 290 correctly predicted images so that the accuracy reached 97%. From the results  of deploying using the website interface using flask, it was found that the classification had an accuracy of above 95% so it was good enough to be used as a prototype for classification engine applications.