Research On The Surface Defects Detection And Recognition For Silicon Steel Sheet

Silicon steel sheet is an indispensable important soft magnetic alloy material for power generation, transmission, electrical, electronic and military industry, mainly used for making all kinds of generator, motor and transformer core. In view of silicon steel sheet surface quality requirement is very strict, the existence of micro defects will seriously affect the performance of the product. Hence, the surface defect detection system of silicon steel sheet required many bigger challenges. Strict product quality inspection can control the production process and improve the process equipment, and then improv the quality of the products. Therefore, the production of high quality silicon steel sheet needs effective detection, so iron and steel enterprises spend a huge sum of money to enhance the detection technology and improve the level of detection. This article analyzed the key technical problems from the perspective of image processing and target detection for the silicon steel sheet surface defects exist in the target detection and recognition system. Furthermore, the solutions were also studied from the image filtering and saliency extraction technology, image segmentation, feature extraction and classification recognition technology respectively. The main research contents and results are listed as follows:(1) The micro surface defect images were studyed, a saliency convex active contour model (SCACM) detection method was proposed to detect the micro silicon steel sheet surface defects under the background clutter and noise interference. In the proposed method, visual saliency extraction was employed to suppress the clutter background for the purpose of highlighting the potential objects. The extracted saliency map was then exploited as a feature, which was fused into a convex energy minimization function of local-based active contour. Meanwhile, a numerical minimization algorithm was introduced to separate the micro surface defects from cluttered background. Experimental results demonstrated that the proposed method presents the well performance for detecting micro surface defects including spot-defect and steel-pit-defect. Even in the cluttered background, the proposed method detected almost all of the micro defects without any false objects.(2) To detect the interesting defect objects for silicon steel strip under oil pollution interference, a new detection method based on saliency linear scanning morphology (SLSM) was proposed. In the proposed method, visual saliency extraction was employed to suppress the clutter background. Meanwhile, a saliency map was obtained for the purpose of highlighting the potential objects. Then, the linear scanning operation was proposed to obtain the region of oil pollution. Finally, the morphology edge processing was proposed to remove the edge of oil pollution interference and the edge of reflective pseudo-defect. Experimental results demonstrated that the proposed method presents the good performance for detecting surface defects including wipe-crack-defect, scratch-defect and small-defect.(3) A data set of silicon steel sheet surface defect images of NEU-Silicon was established to test the stability of the defects feature extraction method. Moreover, a number of samples and defect categories for hot rolled sheet surface defect image dataset NEU-Hot was build. In the data set, the same category defects have biggish difference appearance, while different types of defects have many similarities. Therefore, NEU-Hot image data set is much more challenging than NEU-Silicon image data set, can be more comprehensive verify the performance of feature extraction method.(4) After analyzing the issue of deformation stability for feature extraction methods such as Fourier transform and wavelet transform, the scattering transform convolution network was introduced. In the method, a scattering transform build non-linear invariants representation by cascading wavelet transforms and modulus pooling operators, which average the amplitude of iterated wavelet coefficients. Then, an improved network named the scattering convolution network (SCN) was introduced to build large-scale invariants. SCN was translation invariance and Lipschitz continuity, and can keep the signal energy. Using NNC classifier in NEU-Silicon image data set SCN obtained the average classification accuracy of 95.2%. While using NNC classifier and SVM in NEU-Hot image data set, SCN obtained the average classification accuracy with 97.24% and 97.24% respectively. Experimental results demonstrated that the SCN method presents the excellent performance of defect recognition under the influence of the feature variations of the intra-class changes, the illumination and grayscale changes. Moreover, the experimental results also verify the SCN has a certain application promotion value.(5) In view of the extract feature coding strategy for local binary pattern (LBP) is more sensitive to noise, the neighborhood under the adjacent evaluation local binary patterns (AELBP) was proposed. In the proposed approach, an adjacent evaluation window was constructed to modify the threshold scheme of LBP. Moreover, this adjacent evaluation method was generalized and can be integrated the with existing LBP variants such as completed local binary pattern (CLBP) and local ternary pattern (LTP) to derive new image features against noise. The proposed approaches were compared with the state-of-the-art approaches on Outex and CUReT databases, and evaluated on three challenging databases (i.e. UIUC, UMD and ALOT databases) for texture classification. Experimental results demonstrated that the proposed approaches present a solid power of texture classification under illumination and rotation variations, significant viewpoint changes, and significant large scale challenging conditions. Furthermore, the proposed approaches were more robust against noise and consistently outperformed all the basic approaches in comparison. Moreover, the proposed methods were also tested the application promotion value in NEU-Silicon and NEU-Hot image data set. The experimental results showed that, even in the gaussian noise interference, the proposed AECLBP method could obtain the best average classification accuracy.