The prediction of the quality of fresh fruit (fruit) was the main idea of this project and it was hoped to serve as a recommendation for the selection of fresh fruit for general or industrial use. With difficulty predicting the actual state of fruit uses through observation its external appearance as well as internal quality factors, believing that the Computer A vision could help us solve the problem. Fruit domains are involved in this project they were apple, banana and orange. In general, the development was divided into two parts different tasks or stages, fruit classification and fruit detection or localization. On identify and locate fruit presented in one image or frame, in another data image set was collected for manual annotation and this data was ready to be "put in". an object detection API that has been configured for the Faster R-CNN object detection model as a training channel. A frozen inference graph was trained to use fruit detection. In addition, in order to predict the freshness states of these fruits, data images of two distinguishable freshness states of "Fresh" and "Rotten" were collected. Convolutional A neural network (CNN) architecture was used to construct and train the classifier. Model. In order to achieve the desired performance from the model, the evaluation a analyses were made to improve gradually with use regularization method and transfer learning approach. Finally, the successes of both models, the R-CNN faster detection graph was well trained because it identifies classes in most situations, as with CNN classification The model achieved an overall accuracy of 99.81% due to the use of pre-trained weights from the VGG16 model. Until the current stage, the prototype was of limited use prediction of the quality of fruit domains. It was believed that it could be deployed for the real-world use if improvements and extended development have been made to this prototype system

.