Abstract: This paper discusses how to improve zxing's decoding algorithm for DataMatrix which was used for real-time scanning and recognition in industrial manufacturing field. This paper introduced the principle of DataMatrix and the algorithm steps of zxing, and analyzed the algorithm and pointed out the deficiencies. In view of these problems, a number of optimization measures were proposed, including using a more effective binarization algorithm, taking measures to avoid premature termination of the detection process of the DataMatrix code area, judging the L line according to the black number, and using image convolution to eliminate noise when reading data. Experimental results show that after taking these optimization measures, the success rate of DataMatrix recognition has been greatly improved.