Banana Fruit Classification using Convolutional Neural Network

Abstract

Identifying and recognition is the one of main importance process that has been used in many industries nowadays. To make this process more efficient, the technology such image processing needs to be used. One of industry which is in a lot of demand nowadays is fruit industry. The quality care for fruit plant should be emphasized by the farmers to ensure the fruit that produced is healthy. People usually use guide books or website to recognize the fruit maturity and this consuming a lot of time and hard to bring the accurate result. An automated detection system can help to identify the types of fruit maturity with more efficient. The purpose of this research is to build a banana fruit classification system for identifying whether the banana fruit is raw, ripe or overripe. This project uses 90 images of three different type of banana fruit ripeness. The method used in this research contains several stages, the stages that involved in this research is input image, preprocessing image, segmentation, classification and performance measure. The input images are from the three types of banana fruit ripeness which is raw, ripe and overripe. The dataset is divided into three category which is training, validation and testing. 20 images from each category are used for training and validation while the remaining dataset used for testing process. In preprocessing stage, resizing input image, conversion from RGB image to grayscale image and reducing noise using median filtering method are perform in this stage. In segmentation stage, the images are segment using Sobel edge extraction. In classification stage, the features extraction and identifying process of the banana fruit ripeness is perform in this stage using Resnet-50 classifier. The performance measure is the calculation of mean squared error (MSE), peak signal-to-noise ratio (PSNR), accuracy, and error. The proposed method is capable to determine whether it raw, ripe or overripe. The proposed method also shows that the accuracy and error for raw banana is 92.70% and 7.30% respectively while for ripe banana is 94.32% and 5.68% respectively and lastly, for overripe banana is 95.41% and 4.59% respectively. In conclusion, this project achieved the objective which is to develop a banana fruit classification system using convolutional neural network to determine whether the banana is raw, ripe and overripe.