Steel Surface Quality Assessment And System Based On Image Sequences

With the improvement of global manufacturing level,the downstream manufacturing industry demands higher and higher quality of strip products.As an important part of strip steel product quality inspection,strip steel surface quality evaluation has important guiding significance for customer selection.Based on the problem of determining the surface quality of strip steel products in a certain iron and steel enterprise,this thesis conducts an in-depth study on the evaluation method of strip steel surface quality,and proposes a novel method of strip steel surface quality evaluation.The main research work is as follows:(1)The current research status of strip surface defect detection is summarized,and the existing image feature extraction methods,object detection algorithms and classification algorithms are studied.It lays a theoretical foundation for the design of the follow-up algorithm and the development of the system.(2)In this thesis,a method of defect grade determination based on tf-idf improved multi-feature fusion is proposed.Aiming at the problem of defect grade sample quantity lack of,the TF-IDF algorithm for BOF weighted,and will be weighted BOF and gray histogram features fusion,reusing the thought of image retrieval,the strip surface defect level determination,and combined with artificial correction,to build the variety defect levels sample library,for subsequent Faster-R-CNN implementation defect detection laid the foundation samples.(3)The Faster-R-CNN method for strip surface defect detection is implemented.To solve the problem of low recognition rate of surface analyzer detection,Faster r-cnn network was used to detect surface defects of strip steel.Moreover,experiments were carried out using vgg-16 and resnet-101 as feature extraction networks respectively.The experimental results showed that resnet-101,as feature extraction network,had better detection effect.(4)An image-oriented method for strip surface quality evaluation is presented.First,the strip image sequence is input into the trained detection model,and the defect information of each defect image is obtained.Then,combining the user's historical purchase information and the actual demand of steel production,a feature construction method is proposed to generate new data samples.Finally,the data samples are used to train the classifier,and the test is carried out.Experimental results show that the algorithm can realize the evaluation of unknown strip surface quality,and the effectiveness of the algorithm is verified.(5)Taking the method of strip steel surface quality assessment proposed in this thesis as the core,the system of strip steel surface quality assessment is designed and implemented.Through the input image sequence,the strip steel surface quality assessment is realized.The position,type and quality evaluation of the surface defects on the strip are presented visually.