Optimizing InceptionV3 Hyperparameters Using Multi-Objective Genetic Algorithm for Video Frame Classification.

Automated defect detection in manufacturing plays a crucial role in ensuring product quality and reducing manual inspection costs. Deep learning-based methods, particularly Convolutional Neural Networks (CNNs), have demonstrated significant success in image classification tasks. However, selecting optimal hyperparameters for CNNs remains a challenging problem due to the trade-off between accuracy and computational complexity. This paper proposes a Multi-Objective Genetic Algorithm (MOGA) for optimizing the hyperparameters of an InceptionV3-based CNN to improve defect detection in video frames. The proposed approach finetunes key parameters such as dropout rate, learning rate, number of dense layers, and their respective units, balancing model accuracy and efficiency. The optimized model is trained and validated on a custom defect dataset, and its performance is tested on video frames. Experimental results show that the MOGA-optimized InceptionV3 outperforms baseline configurations, achieving higher defect classification accuracy 99.52% while maintaining lower model complexity. Furthermore, the framework processes video frames, identifying defects in real-time and providing annotated results. The results indicate that optimizing CNN-based defect detection through genetic algorithms not only substantially improves performance but also reduces computational complexity, making it a practical and efficient solution for industrial applications.