Research Of The Key Techniques In Surface Defects Inspection For Strip Based On Machine Vision

Metal strip is the basic raw material of industry,and the surface quality of the metal strip is more and more concerned by enterprises.Surface inspection data can be used not only to analyze the process,but also to provide an important information for the management of finished products.With the rapid increase of market demand,the surface inspection technology based on machine vision has become a hot research topic.At present,the technology market is monopolized by foreign countries,and the technology is still not mature.In this paper,the existing technical problems in the metal strip surface inspection system are researched and the solutions are given after analyzing and summarizing the key technologies.After the prototype was developed,the engineering verification was carried out.Finally the application is successful.In this paper,innovative achievements include:1.To eliminate the non-uniformity caused by variant confounders during the image acquisition,we propose an online correction method by the Gauss-Markov model and the estimation of statistical background.After correction,the images become uniform and have consistent gray level.And then,based on 3? criterion,we design an adaptive defect detection algorithm for variant texture background image corrected,which meets real-time requirements in high data rate.The detection rate and stability of the system are guaranteed by two sequential steps mentioned above.2.To obtain effective segmentation results for the variable texture background and non-significant defect images,we propose a adaptive combinition segmentation method.This method combines 3?Adaptive dual side threshold,residual CHI square detection,auto entropy and based on the least error probability,which is implemented with automatic multi-threading technology.The experimental results show that it is more effective.3.To improve classification rate for the multi-form and non-significant defects,we propose a two level classifier structure,which is composed of the nearest neighbor classifier and bag-of-words(BoW)model based the chaotic features.easily recognizable defect images is classified and filered out by the nearest neighbor classifier in first level,and the orther data will be sent to the second level and classified by BoW model with the chaotic features.4.To get stable and high quality images,several key parts are designed for the surface inspection system: the integrated optical image acquisiton device for multiple cameras,self-locking adjustment device,power line light and light adjustment device for line scan cameras.These patented technologies have been implemented by ourselves.5.To meet the demand of Surface inspection in nonferrous metal processing industry,we develope the surface inspection system for metal with independent intellectual property rights,These systems have been applied and satisfy the requirement of commercial application through continually improving.Finally,we break up the foreign monopoly on this technique.