Research On Detection Algorithm Of Float Glass Defect Based On Multi-feature

Due to the influence of different processing technique and environmental factors on production progress of float glass,there will produce some different types of flaws.By classifying the defects correctly,we can reduce the costs,improve the glass quality and production efficiency.Float glass online detection based on machine vision not only has high detection accuracy and efficiency,but also has reduced unnecessary labor costs and adapted to modern production of glass.The thesis studied serial online detection and recognition algorithms of float glass based on the machine vision.The main work of this thesis contains as follow:(1)Float glass defects image pre-processing.Firstly a subtracting method is adopted to remove stripe background;Secondly median filtering algorithm is selected to remove the interference of noises;Lastly by using gray-level linear transformations we can improve the definition of defect areas' image and reduce visual sense of other areas.(2)Single feature extraction and expansion.The paper extracts three kinds of features: texture feature,shape feature,static feature.Then each feature is expanded to surmounting the disadvantages.The comparisons based on a number of experiments show that expansions are effective and necessary.(3)Defects' multi-features fusion and features selection.Through serial fusion we can get 224-dimensional features.It will take time and unsatisfactory without feature reduction.The paper reduces 224-dimensional features to 17-dimensional features by ReCorre which consists of two steps: removes irrelevant features and then removes redundant features.Experimental result shows that the detection rate of defects by 17-dimensional features is higher than by three kinds of single feature.(4)Float glass defects detection based on support vector machine.When building SVM model to detect glass defects,the kernel function selects Radial Basis Function,the multi-class classification strategy selects one-against-one policy.AMPSO algorithm is selected to find the optimal model parameters in order to overcome the standard SVM' blindness.After AMPSO finds the optimal model parameters the paper builds SVM model to detect glass defects.Experimental result shows that the SVM after choice of the optimization parameters has better performance than standard SVM in detecting float glass defects.