The Initial Study Of Concrete Apparent Quality Evaluation Based On Image Recognition

With the development of economy, the update of conception and the advancing criterion, the external appearance of concrete is becoming significantly increasing, which is regarded as an important guide line in many projects. The demand of a consistently high-quality external appearance finish in products introduces inspection requirements that cannot easily meet by manual inspection. So it is extraordinary to find new methods which are more convenient and credibility.This Paper discusses a new method, which can realize to inspect defects and recognize the various defect types on concrete surface, especially present an evaluating for apparent quality, by the technology of digital image processing and pattern recognition. This dissertation emphasize on the research of feature extraction and neural networks classifier design, which must have credibility and availability to meet the need.According to the characteristic of the various defect types, various methods must be used in feature extraction , pattern recognition and estimate .Such as, the diagnostic parameter of the rusty defect for recognition must be extracted in RGB image; but the diagnostic parameter of the other defects for recognition, suchlike rock pocket , surface voids and holes, can be extracted in grey image; The diagnostic parameter of the image for evaluating must be extracted in binary image. In the course of feature extraction, many methods, including principal components analysis , Matrix principal components analysis , Fisher Linear Discriminant analysis , the method based on entropy and Neural Network, are discussed. Through analyzing and comparing the five methods and considering the characteristic of the concrete digital image, Matrix principal components analysis is chosen to feature extraction for recognition. In the part of pattern recognition, BP Neural Network classifier design is discussed and two BP Neural Network schemes of three-layer topological structure is presented, which are good at recognition of various defect types. In the part of evaluating,according to the characteristic of the various defect types, various methods must be used in changing grey image into binary image; After the binary images are calculated by the means of the predefine morphogenesis operation, the diagnostic parameters for evaluating are extracted and suggested its range, according specialist opinion.Research on the concrete appearance quality evaluating based on the technology of digital image processing and pattern recognition presented in this paper, in a way ,can process digital image and recognize the defect type, which process is digital, technological, systematism and standardized. The last part of the paper, some suggestions are presented.