An Automated Process of Detecting Fresh And Rotten Vegetables by Using Deep Learning

Abstract

Classification of fresh and rotten vegetables is essential in our everyday life as it becomes difficult sometimes to identify earlier whether the vegetables are fully fresh or not. There is a possibility of suffering from severe diseases if defect vegetables are consumed. On the other hand, it will also be difficult to import vegetables in foreign countries if freshness is not found which may result in deterioration of economy. For this, we have has been intended to use deep learning method to detect immediately which fresh and rotten vegetables are fresh and which are rotten. Five vegetables are selected which are- Cucumber, Green capsicum, pointed gourd, cabbage and tomato. These are divided into 10 distinct classes according to fresh and rotten categories. Fresh and rotten cucumber, green capsicum, pointed gourd, cabbage and tomato, along with rotten are kept in the tent box for capturing photos which will be good background for photos as only vegetables can be recognized easily. At first, 1096 and 1016 raw image data of fresh and rotten vegetables are captured respectively. This is done by capturing the photos of fresh and rotten vegetables at various angles. After that these were tilted, rotated at the angles different from the angles of photos existed during capturing and resized to increase more in numbers. Among these, 206 images of fresh and 195 images of rotten vegetables are tested and 830 images of fresh and 786 images of rotten vegetables are trained. Here the four models of deep learning InceptionV3, MobileNetV2, VGG16 and Xception model are applied as they can provide good accuracy. InceptionV3 provides the accuracy of 98.50%, accuracy of VGG-16 is 97.01%, MobileNetV2 provides the accuracy of 99.50% and Xception model has the accuracy of 99.75%. Among these four models, Xception is the most accurate model.