Research On Surface Defect Detection Technology Based On Embedded System

Surface defects detection technology is widely used in fields like steel,textile,paper and printing.It aims at defects detection and defective samples removing.Machine vision-based surface defect detection has a huge advantage over manual manner detection.It ensures high-speed,continuous,accurate online sorting and statistical analysis of quality.The traditional visual inspection system has the shortcomings such as high cost and inconvenient development.Moreover,the surface defect detection algorithm has some problems like lack of generality and poor illumination adaptability.We have done further study on the above issues and achieved achievements as follows:(1)We propose a visual inspection solution based on embedded system.From the aspects of cost,speed,power consumption and detection target,we analyses the hardware structure of the system,the selection of the core processor and so on.The system can meet the system functional requirements,and has advantages of low cost and convenient development.(2)We propose a defect detection method based on Weighted Linear Regression(WLR)and Gaussian Mixture Model(GMM).From the recognition rate,we compare our method with saliency-based defect detection method and the independent component analysis(ICA-based)method.(3)We design a visual inspection software framework based on android system.We analyses the requirements of the system function,divide the functional modules and analyses the relationship between these modules.The multi-thread programming is adopted to ensure the efficient execution.The multi-threading software framework based on Looper-Handler communication mechanism is designed to guarantee the reliability of the system operation.By using NDK(Native Development Kit)technology,the processing speed of the algorithm module is optimized.In this thesis,the performances of the proposed method and the inspection software are tested on the button surface defect.Experiment results show that the proposed method in this paper achieves at least 94% recognition rate on different types of buttons.The method has strong generality and good illumination adaptability.The detection software system can operate stably and smoothly with good interaction experience.