Deep Learning and Computer Vision-Based System for Detecting and Separating Abnormal Bags in Automatic Bagging Machines

This paper presents a novel deep learning and computer vision-based system for detecting and separating abnormal bags within automatic bagging machines, addressing a key challenge in industrial quality control. The core of our approach is the development of a data collection system seamlessly integrated into the production line. This system captures a comprehensive variety of bag images, ensuring a dataset representative of real-world variability. To augment the quantity and quality of our training data, we implement both offline and online data augmentation techniques. For classifying normal and abnormal bags, we design a lightweight deep learning model based on the residual network for deployment on computationally constrained devices. Specifically, we improve the initial convolutional layer by utilizing ghost convolution and implement a reduced channel strategy across the network layers. Additionally, knowledge distillation is employed to refine the model's performance by transferring insights from a fully trained, more complex network. We conduct extensive comparisons with other convolutional neural network models, demonstrating that our proposed model achieves superior performance in classifying bags while maintaining high efficiency. Ablation studies further validate the contribution of each modification to the model's success. Upon deployment, the model demonstrates robust accuracy and operational efficiency in a live production environment. The system provides significant improvements in automatic bagging processes, combining accuracy with practical applicability in industrial settings.