The Steel Surface Defects Region Detection Based On Visual Attention Mechanism

The plate defect detection plays an important role in quality testing, which can make thestaff easily facilitate the process and improve the billet quality. Mostly, in the traditional steeldefect detection, staff identifies the quality when steel body temperature is cooled to a standardone, or extracts images through the camera monitoring. Both methods belong to the artificialdetection, and cost too much time or labor. Therefore, there is great significance to research steeldefect detection system using intelligent detection methods for improving the steel quality.Steel surface defect detection plays an important role in CIMS. It needs searching targets,matching and recognizing when processing surface defects by machine vision detection in thefull map. So it is difficult to achieve real-time online testing. General, as a small region whichsurface defects occupy, it causes the computing resources and storage resources wasted easily.Therefore, it is necessary to introduce visual attention mechanism to shorten the search processthrough determining the initial defect area. And then we can recognize the defect regions toreduce the risk of false alarm and realize real-time accurate detection according to the saliencymap.The paper started at the research background of steel surface defect detection. Afterdiscussing the technical development status at home and abroad, the paper designed a softwaresystem to defect detection based on C/S framework. The paper focused on the steel surfacedefect detection algorithm based on visual attention mechanism and used3000steel images fromcamera to take experiments. Experimental results showed that the proposed software systemcould inspect the plate surface defects effectively and run steadily.