Research on characteristic recognition and quantification of internal powder residue in LPBF porous structure based on image processing

To address the low efficiency and accuracy of residual powder detection in LPBF porous structures, an automated visualization and evaluation method is proposed. Based on CT images, it develops a dual-threshold ensemble grayscale segmentation algorithm on Matlab, integrating morphological processing and U-Net for batch residual powder identification and extraction, and then analyze, compare and recommend the post-processing process based on the residual powder information in the database. Validation shows this method is 20 time more efficient than Image J (12 min vs. 240 min for 1437 images) with accuracy improved to 86.42%-89.21%. Integrated with image quality evaluation and large models, it builds a “detection-recognition-post-processing” system, offering a scalable paradigm for LPBF quality control and defect-property correlation analysis.