Pyramid-based anti-fisheye feature enhancement preprocessing algorithm in torpedo can electrical devices: application in steel rolling process

Abstract The steel manufacturing industry currently urgently needs highly accurate detection algorithms for electrical connection devices to slow down the time and danger of electrical connections to torpedo cans during high-temperature operations. The fisheye effect and fuzzy features of industrial cameras seriously affect accuracy and effectiveness, hindering the widespread application of object detection algorithms in the manufacturing industry. We propose a feature enhancement preprocessing algorithm for torpedo can electrical devices based on the pyramid structure that resists fisheye effects and serves to detect and locate electrical connection devices. With the aid of this preprocessing algorithm, the detection efficiency and accuracy of state-of-the-art (SOTA) object detection models are significantly improved. Experimental validation confirms the superiority of our method over other SOTA methods. With the application of our preprocessing algorithm, the production capacity of the steel plant increased by 31.8%, and material wastage caused by transportation decreased by 10.9%.