Intelligent machine vision solution in manufacturing using deep learning method and artificial intelligence (AI)

Abstract

In this project, we apply the available knowledge of Deep Learning and Artificial Intelligence into building a custom Machine Vision system to make automatic management the quality of the product more efficient. Object Detection and Image Classification can be highly effective for the automatic management of the quality of the product, and the quality of the product that needs to be checked by locating, identifying subjects and classifying. For the test product for building our own custom dataset, we choose floor tile as the test subject. The major steps in our proposal are detecting and classifying the test subjects in the states of defective and non-defective, then sending a signal for the conveyor belt to distinguish products’ error states and non-error. We test our custom model by training both Object Detection and Image Classification models, the SSD MobileNet model and the InceptionV3 model respectively. The obtained results show the differences of both models in terms of Accuracy and FPS. The SSD MobileNet model used for Object Detection has higher FPS than the InceptionV3 for Image Classification, while the InceptionV3 has higher accuracy. Further research among examined algorithms and the structure have been presented in this thesis.