mariam-diab / Automated-Detection-of-Defects-on-Metal-Surfaces-using-Deep-Learning-Techniques-and-Vision-Transform

The project focuses on utilizing deep learning and transformers to detect defects in metals. Tensorflow has been employed to design the model architecture. Among the various approaches explored, the Inception model, a custom CNN model, and the VIT feature extractor have yielded the most promising results.

Automated Detection of Defects on Metal Surfaces using Deep Learning Techniques and Vision Transform

This project utilizes machine and deep learning techniques to automate the detection of defects on metal surfaces in industrial products. The goal of this project is to classify various common metal defects. The dataset can be found here https://www.kaggle.com/datasets/toqaalaaawad/metal-surfaces-defects
ِAll of the work can be found in the attached PDF report file
The table shows the types of defects we're working on

Crease	Crescent gap	Inclusion	Oil Spot
Punching hole	Rolled in Scale	Rolled pit	Scratches
Silk spot	Waist folding	Water spot	Welding line

This repository serves as a comprehensive record of all the steps taken to improve the accuracy of the system.

The repository includes the following:

• Categorical classification between two classes using Sigmoid activation.
• Categorical classification among 10 defect classes using a custom CNN model with softmax.
• Categorical classification among 10 defect classes using Inception and SGD.
• Categorical classification among 11 defect classes using a custom CNN model with softmax.
• Categorical classification among 11 defect classes using Inception and SGD.
• Single detection of a defect by training the Inception model on the image annotations.
• Multiple detection of defects by training the Inception model on the image annotations.
• Using VIT as a feature extractor instead of the pretrained model.

Summary of the most promising results:

Multiple detection of defects using inception:

The model architecture:

Multiple defect detection model architecture

The model results:

The accuracy and the validation metrics

Single detection of a defect using inception (Trained on annotations):

The model architecture:

Single defect detection model architecture

The model results:

The accuracy and the validation metrics

Single detection of a defect using custom CNN model

The model architecture:

Multiple defect detection model architecture

The model results:

The accuracy and the validation metrics

Single detection of a defect using Inception with SVM:

The model architecture:

Multiple defect detection model architecture

The model results:

The accuracy and the validation metrics