Real-Time Detection of Defective Products of a Tortilla Machine Production Line Using TensorFlow Object Detection API and OpenCV

This work focuses on the real-time detection of defective products on a production line of a Roti (Flatbread) maker machine. In this research work, an object detection model, SSD-MobileNet-v2 has been used and fine-tuned through transfer learning with the help of a custom dataset. The model is trained for 10,000 steps and demonstrated superior performance, achieving mAP and recall values of 0.9278 and 0.9379 respectively. This defect identification in real-time leads to huge cost savings and human interventions. This is not only beneficial for the food production industry but also holds potential for broader applications in large-scale industries including FMCG products. The integration of these models into production lines offers invaluable tools for ensuring high standards of quality and safety across various sectors.