Optimizing Inception Architectures for Automated Quality Control in Binary Classification

In computer vision, within convolutional neural network architecture, Inception algorithms play a crucial role for image classification tasks. This study focuses on optimizing Inception architectures for binary classification, particularly for automating quality control processes to simplify control phases and improve accuracy. Conventional quality control methods often rely on manual inspection, which can be time-consuming and prone to human error. The optimization process involves careful consideration of transfer learning methods, the benefits of incorporating fine-tuning and other regularization techniques such as data augmentation, choosing the right values for dropout layers and learning rates will show the best results in terms of accuracy and efficiency. The examination of three different Inception algorithms, Inception v3, Inception Resnet v2, and Xception, reveals that Xception achieves higher validation accuracy 99.72% compared with Inception Resnet V2 and Inception V3 in larger datasets. In the other hand for smaller dataset, Inception V3 achieves higher validation accuracy 99.69% compared with Inception Resnet V2 and Xception. The decision-making regarding the use of these algorithms should be guided by the specific use cases, as each algorithm presents distinct strengths suitable for quality control applications. Keywords: CNN, inception algorithms, transfer learning, quality control, image classification