Research On The Technology To Hot Flatness Defect Pattern Recognition

Hot-rolled steel is an important one for a variety of steel products. With the development ofthe hot-rolled sheet in all aspects of application, the requirements for the products are moreand more strict. Users are not only value the quality of steel, but also take more attenion to theappearance of the steel. Sometimes, the appearance of hot-rolled steel sheet is a veryimportant indicator for the evaluation of hot-rolled steel sheet. The customers will not acceptthe products with quality and appearance not reaching the demands. Therefore, we should notonly concern about the performance and quality of the products, but also the appearance. Butduring the process of hot-rolled, products may not totally meet the customers’ requirementsowing to the factors of high temperature, high pressure, high speed and hardware. The surfacedefects of steel plates will be inevitable and run through the process of hot-rolled. Theyclosely relevant with the appearance of follow-up forming devices, but also the follow-upprocessing effects and costs.In recent years, for hot-rolled steel plates getting thinner andthinner and the increasingly high precision of the shape accuracy, it has become particularlyimportant to extensive research for the defects of hot rolled steel sheet.The defect recognition of hot-rolled steel surface is an important applications of patternrecognition, in this article, two aspects of research work are studied on the classification andidentification of the hot-rolled steel sheet. Hot-rolled steel sheet classification of commonsurface defects are reviewed in detail, and some causes of defects are analyzed. Meanwhile, anumber of hot-rolled steel plate to prevent the occurrence of defects are discussed. Finally,two methods were adopted, based on wavelet transform coefficients and wavelet packet,combined BP neural network methods, the defects are classified and identified. Two wayswere compared, also they have a higher recognition rate compared to the traditional methodthat image feature extraction by frequency domain transformation,and the extracted featurevector dimension is relatively high, not easy to achieve, but with the method of waveletpacket image feature extraction can accurately reflect the image features of the imageinformation, meanwhile, the algorithm to extract the energy feature vector of image featurescan be a better description of defects, including low and high informations of the images andhave robust noise immunity and good scalability, also reflects the superiority of this algorithm.