Real-Time Bottle Classification Using MobileNetV2: An Evaluation of Beer, Plastic, Soda, Water, and Wine Bottle

In several fields, especially retail, recycling, and trash management, object classification is important. This paper investigates five types of bottles: Beer Bottles, Plastic Bottles, Soda Bottles, Water Bottles, and Wine Bottles using a convolutional neural network (CNN). Based on a 25,000-image balanced dataset, the paper evaluates MobileNetV2's F1 score, accuracy, precision, and recall. Plastic bottles show almost perfect accuracy and recall of 99%, so the model attained an astounding general accuracy of 97%. Additionally showing great accuracy rates of 97% and 98%, respectively were water and beer bottles. The model clearly distinguishes between bottle types despite small misclassifications, including Soda Bottles being sometimes confused with Beer Bottles. Good model tuning with low over fitting is shown by the study of training and validation losses. The model's resilience even further is confirmed by the close alignment of the training and validation accuracy. The confusion matrix points up various misclassifications, which offer information for the next developments. This work highlights MobileNetV2's great fit for real-time bottle categorization uses, therefore providing major advantages for retail and recycling sectors of industry automation. Future research will concentrate on improving the model to solve found misclassification problems and raise its practical relevance.