Machine vision is one of the cutting-edge technologies that can assist human operators in tasks such as bottle-fill level detection and classification, resulting in increased efficiency in the bottling industry. Although pre-trained models such as the MobileNet, ResNet-50, and VGG-19 for bottle-fill level detection and classification exist, their accuracy is dependent on the similarity of their trained data to the application domain. As a result, this paper describes how to create a sequential neural network model for bottle-fill level detection and classification in Python 3.8.3. The proposed model is evaluated and compared to the MobileNet, ResNet-50, and VGG-19 models in a multiclass problem (correctly filled, overfilled, and underfilled). Furthermore, a confusion matrix was used to assess the performance of the proposed model in the correctly filled, overfilled, and underfilled categories of filled bottles. In comparison to the MobileNet, ResNet50, and VGG-19 models, the proposed model had a training and testing dataset accuracy of 98%, while the MobileNet had 73%, ResNet50 had 76%, and VGG-19 had 75%. The accuracy of the confusion matrix on 40 sample sizes for each class of the filled level was 97%. Finally, in the application domain, the proposed model outperforms the MobileNet, ResNet50, and VGG-19 models. As a result, the method used in the neural network layer structuring of the sequential neural network model should be considered a viable alternative in similar applications.