Coin Detection And Defect Recognition Based On Machine Vision

In July 2018,the People's Bank of China issued the industry standard“Not Circulating RMB Coins” according to the features of RMB coin which is not suitable for circulation.The coins' defects are mainly divided into the following categories: specifications,stains,abrasion,discoloration,deformation,holes and cracks.The features of coins that are not suitable for circulation are specifically manifested as at least one of above-mentioned defects which do not meet the standard.Manual sorting of such coins that are not suitable for circulation is not only inefficient,but because of the subjective factors of workers,it is impossible to unify the standards.In response to such situations,the paper builds a coin detection and defect identification system to improve the detection accuracy of coins that are not suitable for circulation and unify the recovery standards.The research content of the thesis mainly includes:The paper builds a coin image acquisition device,which mainly includes three color industrial cameras,three coaxial light sources,fixed baffles and a weighing platform.The paper proposes a method for extracting coin targets in an image.On the basis of preprocessed image,the foreground and background of the image are segmented through image superposition and morphological operations to achieve the extraction of coin targets.Based on the segmented image,we calculate the minimum outer rectangle of the coin and obtain the diameter of the coin,then combine the quality of the coin to distinguish the identification of the coin currency.At the same time,we compare the size of the long and short axes and complete the deformation detection and specification judgment of the coin.The thesis classifies holes and cracks into one class for identification,detects and distinguishes holes and cracks by contour tracking and flood filling.The thesis classifies stains and abrasion into one class for identification and determines the threshold of image segmentation by the double peak method to achieve the positioning of the defect area in the coin image.Combined with the pixel variance in the defect area,stains,abrasion detection and identification of coins are completed.After the conversion of coin image from RGB color space to HSV color space,the color change detection of the coin is completed according to the comparison of H and S components.The thickness of the edge of the coin is calculated based on the side image of the coin to determine whether the coin is deformed.Finally,according to the demand,we count the defect information that needs to be displayed and write the interface program based on MFC.After testing,the system has a qualified detection accuracy rate of 99.8% for coins,and an accuracy rate of more than 90% for various types of defects.The average detection speed is about 50 ms,which can meet the requirements of real-time detection.