| Optical character recognition (OCR) categorizes text in images, such as license plate numbers. However, the industrial sector avoids OCR due to model training limitations with specific data, complicating the handling of incoming images. Industrial settings introduce noise to images, impacting quality and classification. This paper seeks to boost machine learning by constructing models with diverse datasets. Focusing on deteriorated serial plates in ship engine rooms, the study employed color inversion and two You Only Look Once (YOLOv5) models for object detection. Subsequently, the Residual Network 152 (ResNet-152) model classified the alphanumeric characters on the plates, achieving 90% accuracy in alphanumeric character classification on deteriorated serial plates. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.