Enhancement of Image Classification Using Transfer Learning and GAN-Based Synthetic Data Augmentation

Plastic bottle recycling has a crucial role in environmental degradation and protection. Position background should be the same to classify plastic bottles on conveyor belt. The manual detection of is time consuming leads human error. Hence, automatic classification using deep learning techniques can assist with more accurate results reduce cost. To achieve considerably good result DL model, we need large volume data train. We propose GAN-based model generate synthetic images similar original. improve image synthesis quality less training decrease chances mode collapse, modified lightweight-GAN which consists generator discriminator an auto-encoding feature capture essential parts input encourage produce wide range real data. Then newly designed weighted average ensemble based two pre-trained models, inceptionV3 xception, transparent obtains improved accuracy 99.06%.