Glass surfaces are widely employed in various applications, including glass surfaces built in various automobiles, electronics, construction, and household appliances. If the glass surfaces are not properly produced, they are more likely to develop scratches or deep surface cracks. In this research work, scratch detection on the glass surfaces using a convolutional neural network (CNN) is carried out. Conventional scratch detection methods frequently rely on manual examination, which is challenging, time-consuming, and error-prone, demanding an automated procedure to handle the problem. This research illustrates the applications of CNN and Visual Geometry Group 19 (VGG19) to identify scratches on glass surfaces reliably. The VGG19 is observed to be efficient in scratch detection on glass surfaces due to its well-suited architecture. The process initiates with the data set and its pre-processing. In the second step, augmentation is applied, followed by feature extraction using the CNN + VGG approach; thereafter, the model is trained after which the model identifies scratch and non-scratch on the glass surface images. In terms of precision and accuracy generated by the model, it gives around 86.3% accuracy, 81.13% precision, and a recall value of 86.00%. It is observed that by utilizing the proposed machine learning method, the system can improve the overall quality control process in glass manufacturing and increase glass surface quality along with its life, performance, and efficiency.