Deep Learning Based Metallic Surface Detection
Metallic surface defect detection is an essential and necessary process to control the qualities of industrial products. Existing defect datasets are available but there is need to deploy detection model for steel defect detection and classification. Traditional detection approaches are poor in both
2025-06-28 16:31:05 - Adil Khan
Deep Learning Based Metallic Surface Detection
Project Area of Specialization Artificial IntelligenceProject SummaryMetallic surface defect detection is an essential and necessary process to control the qualities of industrial products. Existing defect datasets are available but there is need to deploy detection model for steel defect detection and classification. Traditional detection approaches are poor in both efficiency and accuracy for the complex real-world environment. To address this problem, we contribute in applying a real-time detection model for large-scale metallic surface defect detection, with high precision and accuracy. We have selected the Convolutional Neural Network based YOLO Darknet model to perform this task.
In this project, two datasets named as GC10-DET and NEU-DET will be used which contain images of metallic surface defects collected in real industry situations. GC10-DET dataset includes 10 types of surface defects, i.e., punching (Pu), weld line (Wl), crescent gap (Cg), water spot (Ws), oil spot (Os), silk spot (Ss), inclusion (In), rolled pit (Rp), crease (Cr), waist folding (Wf). Whereas, NEU-DET dataset contains 6 kinds of typical surface defects of the hot-rolled steel strip, i.e., rolled-in scale (RS), patches (Pa), crazing (Cr), pitted surface (PS), inclusion (In) and scratches (Sc). Both datasets contain grey-scaled images.
Project ObjectivesOur goal is to train a classifier which should be fast enough to provide real time surface defect detection. Traditionally, filters are applied to an image at multiple locations and high scoring locations are called detections. But we are using a different approach here. We apply a neural network to an image which divides image into different regions and then bounding boxes are predicted for each region. A very similar approach is proposed by a real time detector named You Only Look Once (YOLO) which is our inspiration to perform detection in this way.
Project Implementation MethodThis project deals with the task of metallic surface defect detection from the imagery of defects on surfaces. For this, we gained the basic knowledge of Machine Learning and Deep Learning. The prime focus was given to the Convolutional Neural Network to learn how it is used to build an image classifier. Various improvements techniques such as augmentation, dropout and fine-tuning were implemented.
In this project, we have selected open-source GC10-DET & NEU-DET datasets. The primary task in this thesis is the annotation of these datasets. Annotations on the images of this dataset will be generated using BBOX Label Tool. The secondary task is to train a model to detect defects. For this, we have selected YOLO Darknet model that has the characteristic of getting trained on custom dataset really fast, giving a good PR score. The future work is to use the already learnt improvement techniques along with the learning of new ones to train the YOLO Darknet on our own dataset and get better results.
The benefits of this project are:
- Detect the defects on metallic surface which can lead to industrial equipment failure if missed by an human eye
- Classify the defected areas according to their classes like rolled-in scale, patches, crazing, pitted surfaces, inclusion, scratches, crescent line, welding line, water spot, silk spot, oil spot, crease, punching, waist folding and rolled pit
- Take the precautionary measure well in time
- Control the qualities of industrial products
- Prevent the failure of machine
- ML-based predictive fault detection saves cost and time
- It is also beneficial for industries like the airline industry as airlines need to be extremely efficient in flight operation.
The final system will be designed to automatically detect metallic defects in real time. The trained model built will be fed with either processed images or processed video stream of metallic surface to detect the defects in real-time.
The Operating System environment used will be Linux. The real-time detection speed will be achieved using CUDA Driver of GPU System; and defect detection of video stream will be obtained using OPENCV having CUDA Support.
Final Deliverable of the Project Software SystemCore Industry ManufacturingOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Industry, Innovation and Infrastructure, Sustainable Cities and CommunitiesRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 0 | |||
| NVIDIA GTX 1660 ti | Equipment | 0 | 40000 | 0 |