Performance Comparison of Deep Learning Techniques for Classification of Fruits as Fresh and Rotten
J N V D Tanuia Nerella, Vamsi Krishna Nippulapalli, Srivani Nancharla, Lakshmi Priya Vellanki, Pallikonda Sarah Suhasini
Abstract
The agricultural industry has enormous need for detecting rotten fruits. If the defective fruits are not identified early, the affected fruit may also contaminate healthy fruits. Farmers find it tedious to classify fruits manually into clean and rotten varieties. The manual processes are ineffective and time consuming. Computer vision and Deep learning methods are very useful for automatic segregation systems, reducing the human effort, cost and time. From the fruit images features are extracted in the convolutional layers of Convolutional Neural Network (CNN) and for classifying the images as fresh and rotten, softmax function is used. Different state-of-art Deep learning models including ResNet50, MobileNetV2, VGG 16 and Inception V3 are trained, tested on fruit dataset available from Kaggle website and compared to find the robust model for fruit grading. An accuracy of 97.1 % is obtained with Inception V3model, which is found to be superior over other models.