Bubble Detection in Glass Manufacturing Images Using Generative Adversarial Networks, Filters and Channel Fusion

With the increasing production of glassware products, the detection of bubble defects has been of vital importance. The manual inspection of glass bubble defects is considered to be tedious and inefficient way due to the increasing volume of images, and the high probability of human error. Computer vision-based methods provide us with a platform for automating the bubble defect detection process which can overcome the disadvantages associated with manual inspection thereby significantly reducing the cost and improving the quality. To address these issues, we propose an integrated deep learning (DL) based bubble detection algorithm, in which an image data set is prepared using a Generative Adversarial Network (GAN). The proposed algorithm exploits the Information-Preserving Feature Aggregation (IPFA) module for achieving semantic feature extraction by maintaining the small defects’ internal features. To weed out irrelevant information due to fusion, the proposed research introduces the Conflict Information Suppression Feature Fusion Module (CSFM) to further advance the component combination methodology, the Fine-Grained Conglomeration Module (FGAM) is employed to facilitate cooperation among feature maps at various levels. This approach mitigates the generation of conflicting information arising from erroneous features. The algorithm improved performance with an accuracy rate of 0.677 and a recall rate of 0.716 with a precision value of 0.638.

@article{Ahmed2024,
  title     = {Bubble Detection in Glass Manufacturing Images Using Generative Adversarial Networks, Filters and Channel Fusion},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {9},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Md Ezaz Ahmed and Mohammad Khalid Imam Rahmani and Surbhi Bhatia Khan},
  doi       = {10.14569/IJACSA.2024.0150961},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150961}
}