The Research On Strip Surface Defects Detection Based On Visual Attention Mechanism

As an important part of iron and steel products,the strip steel is also one of the necessary raw materials in the modern industrial development.However,the surface defects will seriously deteriorate the final performance of strip steel products.Therefore,the rapid and accurate detection of strip surface defects is of great value and significance.Machine vision for surface defects detection is an efficient and reliable nondestructive detecting technology,which has attracted widely attention of researchers.The conventional defect detection method implemented on the practical application is of poor real-time and low detection accuracy.Thus,this thesis presents two defect detect method based on Visual attention mechanism model: Gabor wavelet based and spectrum analysis based defect saliency detection method.The contents and results of this paper are as follows:Firstly,this thesis describes the visual attention mechanism model,and a model of strip steel surface defect detection is proposed by employing the traditional visual attention model.Secondly,The Gabor wavelet defect detection method is presented because it is consistent with the human visual attention mechanism with a good choice of direction and scale.To reduce the adjusted parameters for the traditional Gabor wavelet and enhance the real-time detection performance,a Gabor filter optimization method based on the complex differential evolution is proposed.Some experimental results show that detection efficiency has been greatly improved.In order to improve the real-time performance and reliability of the detection algorithm,a defect saliency detection method based on spectrum analysis is proposed,and the amplitude and phase information for image are extracted.Thus,the real-time performance for the surface defect detection is enhanced by using the high efficient computation characteristics of the Fourier transform.Due to the influence of external uniform illumination,the image captured by the camera is prone to local reflection.In order to resolve this problem,an improved Homomorphic filtering method is proposed to remove the reflection area.In view of the problem of obvious image defect,a weighted Mahalanobis distance threshold method is proposed based on the traditional Mahalanobis distance threshold and Pixel weighted.The experimental results show that the defect detection rate is improved.In order to improve the reliability of the experiments,this thesis has construct the laboratory simulation test platform and interactive software interface.The experimental results show that the CoDE-Gabor has a better detection rate and AAP has a good real-time.CoDE-Gabor is suitable for the industry with high quality requirements,such as precision devices and aerospace materials.In addition,AAP is suitable for rapid detection.