Mask R-CNN-based sample piece defect identification and calibration method

• Main Authors: BU YUFENG, YAN JUN, DU HONGZE, ZHOU RUONAN, ZHAO CHUNYU, XU QI, JIANG LIZHE
• Format: Patent
• Language: Chinese; English
• Published: 22.03.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The invention discloses a Mask R-CNN-based sample piece defect identification and calibration method, and relates to the technical field of defect identification and calibration. The method specifically comprises the steps of obtaining a digital image of a sample; obtaining a network extraction feature map after processing, inputting the network extraction feature map to the RPN, and generating a suggestion box; converting the generated suggestion box into the same dimension, respectively sending the suggestion box into a classification network and a regression network, carrying out defect identification, generating a corresponding mask, and completing instance segmentation; training and testing the model to finally obtain a trained model; utilizing the trained model to identify and calibrate internal defects of the sample piece; and performing permutation and combination on the obtained masks according to the sequence of the original digital images to generate a three-dimensional model containing defect cali