Design And Implementation Of The Semantic Segmentation-Based Steel Surface Defect Detection System

China’s steel production level has always been at the forefront of the world.For a long time,the steel industry,as a pillar industry in China’s basic industries,has played an irreplaceable and important role in the rapid development of China’s economy and society.However,the imperfection of the production process and the backwardness of the production equipment may lead to defects on the steel surface during the production process.Surface defects affect the appearance of the product,but also to a certain extent reduce the corrosion resistance and fatigue strength of steel products.Therefore,the intelligent and efficient detection of steel surface defects can have a profound impact on the production efficiency of steel companies and even the entire social and economic construction.With the growing theories of image processing,machine vision,and neural networks,steel surface defect detection technology has gradually become a focus of research in industry and academia.To address the above-mentioned problems,after analyzing the current situation of surface defect detection,we study the surface defect detection,image semantic segmentation-related technologies and the application of image semantic segmentation to steel surface defect detection.Based on the above study,the feature pyramid network(FPN)is improved and a feature pyramid network model based on enhanced perceptual field is constructed to enhance the feature extraction of steel surface defects by feature cascading and improved perceptual field module for the problems such as small overall target and blurred segmentation of steel surface defects.In addition,a steel surface defect detection system is designed and implemented,in which the data pre-processing module mainly collects real-time inspection images and images for unified standardization;the real-time marking module detects defects for real-time steel surface images and realizes defect classification and marking to provide data basis for subsequent modules such as alarm and quality analysis;the information printing module provides real-time inspection results.The quality analysis module realizes the quality statistics and analysis of historical production data.Through experiments and tests,the effectiveness of the improved feature pyramid network is verified in steel surface defect detection,while the system achieves data acquisition,defect marking,analysis,alarming,display and other related functions,meeting the basic needs of users.