About Project
This system for classifying good and defect weld uses the Pytorch pre-trained model SqueezeNet on the TIG aluminum 5083 dataset and it reaches a test accuracy equal to 99.5%. There were several experiments, and the models were trained using 6-classes, 4-classes, and 2-classes tests with the highest test accuracies of 71%, 83%, and 99.5%, respectively. Finally, the 2 models were deployed using Rest API with flask, the first one is to detect if the image is good to weld or defect and the second one is to classify the defect in the image with test accuracy equal to 99%.
was the pre-trained model: pytorch squeezenet model with 99.2% test accuracy on TIG Aluminum 5083 dataset.
https://www.kaggle.com/datasets/danielbacioiu/tig-aluminium-5083