Identifying carpet defects

How Computer Vision Solved the Problem

The traditional method is based on time-consuming visual inspection by humans, which can result in inspectors inadvertently missing some defective products.

In recent years, with the help of digital image processing techniques and computer vision, automated fabric defect inspection has become popular and has gradually replaced the traditional method. Cameras are used to inspect carpets during production, and computer vision techniques look for defects in the products and report them. (Figure 2)

About the Project

To use computer vision to detect defects in the carpet industry, it is necessary to perform the following steps:

1- Collect sample pictures of defects

2- Identify and label all the defects on the images (Figure 3), A process known as annotating

3- Train the model with collected sample defects

At this stage, we have a computer vision model capable of  identifying, categorizing, and reporting fabric defects automatically.

Figure 3. annotating carpet defect (a) hole defect (b) color defect

For this project, we will use the AIEX platform to obtain the desired computer vision model. First, we upload all the sample images of carpet defects to our new project, then we annotate them one by one to indicate what the model should look for. The last step is to choose the type of training we want to use and start training our model. After training concluded, we have a model ready to detect carpet defects. We can test it by feeding the model with new images and inspecting the outcome (Figure 4).