Research On Information Enhancement And Extraction Of Direct Part Marking Image With Low Quality On Metal Surface

In the industrial field,Direct Part Marking(DPM)is used to record the material information,processing information,quality information,and circulation information of industrial product parts,thus realizing the traceability of industrial products,which is very important for industrial products.Full lifecycle management is of great significance.Generally,DPM chooses the Data Matrix barcode with high information density and good error correction performance as the identification carrier.Due to the advantages of non-contact,high processing flexibility and environmental protection,the laser marking method is usually used as the processing method for DPM barcodes.The first step in reading the DPM barcode is to capture the barcode image.However,due to the influence of lighting conditions,substrate surface texture,environmental factors,etc,the quality of the obtained DPM barcode images is low,and many images can not be successfully read.To solve this problem,the enhancement and extraction technology of low-quality DPM barcode image information on the metal surface are researched,mainly from the following aspects.The deblurring algorithm of fuzzy DPM bar code images is studied.Aiming at the problem that the dynamic range of grayscale distribution of DPM barcode images is small,a deblurring algorithm of low dynamic barcode image based on binary features and weighted L0 gradient prior is proposed.Because the fuzzification process will reduce the distance between the two peaks of image grayscale histogram in most binarization cases,the binarization feature can make the estimated image a binary image,and the weighted L0 gradient prior can make the estimated image gradients sparse.On this basis,the numerical optimization algorithm based on the semi-quadratic splitting scheme is adopted to solve the optimization model.Experiments show that proposed method is more effective in visual quality and objective numerical measurement than the existing deblurring methods.The binarization algorithm of low-quality DPM barcode images based on a semantic segmentation network is studied.Learning from the advantages of a semantic segmentation network in processing document images with uneven gray levels,this paper introduces a semantic segmentation network into the binarization of DPM barcode images with uneven gray levels.Three semantic segmentation network models are established,namely U-Net,U-NetVGG16,and U-Net++models.The data sets for model training and testing are self-built data sets.Experiments show the binarization effect of the semantic segmentation network is obviously better than that of the traditional threshold segmentation algorithm.This paper studies the information extraction algorithm of low-quality DPM barcode images.Aiming at the uneven grayscale of low-quality DPM barcode images,and information extraction algorithm for DPM barcode images based on local adjacent modules is proposed.Firstly,the obtained DPM barcode image is tilted and corrected so that the L-shaped solid line boundary can be correctly located at the lower-left corner of the image.Then,the grid is divided to realize the positioning of each module.Next,a neural network model is established to locally learn the color attributes of two adjacent modules.They are marked according to whether the color attributes are the same.The Kuhn-Munkres graph theory algorithm is used to correct the barcode edge image.Finally,the DPM barcode symbol is reconstructed based on the barcode edge image.Experimental results show that the algorithm can extract information from lowquality DPM barcode images,and the decoding success rate is high.