Towards the steel plate defect detection: Multidimensional feature information extraction and fusion

Zhiqiang Hao,Zhigang Wang,Dongxu Bai,Shiyang Zhou

DOI: https://doi.org/10.1002/cpe.6384

2021-05-29

Concurrency and Computation: Practice and Experience

Abstract:<p>Product surface quality inspection based on machine vision has been paid more and more attention in modern industrial production. Multidimensional feature information fusion also plays an important role in the detection rate and accuracy of steel plate defects. Based on the machine vision based defect detection of steel plate surface, this article mainly studies the collection, extraction and fusion of multidimensional feature information. The visual imaging system, the surface defect detection technology based on vision, the depth camera and the light source type are briefly reviewed. At the same time, the roadmap and related methods of multidimensional feature information acquisition system are studied. Finally, the multidimensional fusion defect detection technology is prospected and summarized.</p>

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the extraction and fusion of multi - dimensional feature information in the surface defect detection of steel plates. Specifically, traditional contact - based detection methods can no longer meet the requirements of modern industrial production for product quality, especially in harsh production environments such as steel mills. Therefore, surface defect detection technology based on machine vision has gradually become a research hotspot.

### Main problems include:
1. **Low classification rate and low recognition rate**: Some existing methods perform poorly in the classification and recognition rates of individual defect targets.
2. **Interfering factors such as illumination changes**: These factors affect the stability and robustness of the recognition system.
3. **Insufficient adaptability when the number of samples changes**: As the number of samples changes, it is difficult for the system to maintain good performance.
4. **Neglect of multi - dimensional defect feature information**: Many methods fail to effectively extract and fuse multi - dimensional defect feature information.

### Goals of the paper:
By researching and developing a method capable of collecting, extracting, and fusing multi - dimensional feature information, improve the accuracy and efficiency of surface defect detection of steel plates. Specifically, this paper aims to:
- Collect and extract multi - dimensional feature information of surface defects on steel plates.
- Fuse this multi - dimensional feature information to improve the accuracy of defect detection.
- Propose a multi - dimensional feature information fusion detection technology based on machine vision to deal with the influence of illumination changes and other interfering factors.

### Key points of the solution:
1. **Multi - dimensional feature information collection system**: Build a visual imaging system that can obtain multi - dimensional feature information, including light source selection, lens configuration, and industrial camera selection.
2. **Image pre - processing and information fusion**: Pre - process the collected original images (such as enhancement, segmentation), then analyze the relationships between images of different dimensions, and finally achieve the fusion of multi - dimensional feature information.
3. **Multi - dimensional feature information fusion**: Map multi - dimensional images into high - dimensional defect images to obtain more accurate spatial position relationships.

Through the above methods, this paper expects to make breakthroughs in the surface defect detection of steel plates and provide technical support for intelligent manufacturing.