Study Of Image Feature Extraction Algorithm Based On Lighting Control Of Steel Plate Detection

In segmental machining or seamless assembling process, the requirement of inspection foredge and surface flatness of large size medium and heavy plate is very strict. So, the detectionaccuracy of strip dimension and shape plays a very important role in the development of theindustry. Particularly in recent years, with continuous development of intelligent detectiontechnology, machine vision has been widely applied in precision measurement of the mediumand heavy plate embodies. In the image recognition and detection system based on machinevision, illuminating quality is important for accuracy of inspection. However, in the existingdetection system, intensity of light source only by artificial subjective judgment for lightcondition is adjusted by worker. Therefore, it has not satisfied the demand of detection imagequality. Thus, the analysis of laser light source characteristic and theory of adaptive controlsystem is significant for on-line detection technology of medium and heavy plate.Based on the review of domestic and international research and application of strip shapemeasurement, this thesis put emphasis on the fundamental of imaging mathematical model tolighting system in detection system based on machine vision, design of the control system andsimulation of the fuzzy PID control algorithm to image definition, extraction algorithm based onclassification of light spot center in interference of water drop, and construction of measuringcontrol system.(1) Firstly, the background and significance of this thesis was introduced. Then, the researchstatus of domestic and overseas strip shape detection was discussed. And then, the developmentand application of image feature extraction in steel plate inspection field was summarized.What’s more, the application condition of adaptive control in strip detection, especially in lightsource, was in detail discussed. In addition, the advantages and disadvantages of image featureextraction in detection of steel plate were pointed out.(2) Based on machine visional detection of steel plate, the theoretical relationship betweenpoint laser intensity and output voltage of COMS camera diode was deduced, by buildingimaging mathematical mode. And then, the functional relationship between voltage and image sharpness was calculated, according to the relevant data. Finally, the model parameters weredetermined and virtually realized by experimental methods.(3) On the basis of direct digital control (DDC) system, control system was described; thecontrolled object mathematical mode and transfer function were calculated in accordance withlighting control system control idea. To ensure the definition of collected images in this system,the image quality was controlled with a fuzzy PID algorithm. Furthermore, the fuzzy PID wassimulated and analyzed with Simulink module of Matlab software platform.(4) The fourth chapter presents the characteristics of image acquired, analysis of noise,adaptive threshold and wavelet transform of two preprocessing algorithm of image, andcomparison of advantages and disadvantages for traditional spot extraction algorithm, in thislighting control system of steel plate detection. In order to improve the accuracy of detectionsystem, A extraction algorithm based on classification was put forward, according to imagingcharacteristics of steel plate surface in interference of water drop. And then, this characteristicalgorithm was carried out with Matlab software platform.(5) A steel plate detection system based on the image acquiring was constructed，and thecontrol parameters of image acquiring interface constructed with VC++software platform wasintercalated. Then, precision of light spot center extraction on steel plate surface with andwithout interference of water drop was analyzed and compared. In the experiments, the mainwork is to classify the characteristics of images in interference of water drop. Particularly, inthree kinds of images with serious disturbance of water drop, the light spot center coordinateswere extracted by classified algorithm. Finally, the results show that the detected accuracy isimproved by16.1464pixels, compared with those conventional methods.