Defective Fruit Classification using Variations of GAN for Augmentation

Abstract

Identification and segregation of defective fruits from healthy ones is an important task in the fruit processing industry. In this research paper, we showcase a method for defective lemon fruit classification using different versions of Generative Adversarial Networks (GANs) and Transfer Learning. The algorithm begins with preprocessing the lemon images followed by data augmentation using GANs. GANs generated different versions of original lemon images, which further helped in increasing the size of training data which is required for improving the classification accuracy. After this, all the original and augmented images used as training dataset, which has been utilized by pre-trained Convolutional Networks (CNNs) model where fine-tuning helped in classifying test images. Here, the Lemons Quality Control Dataset was used as the base dataset for conducting all experiments throughout this work.