Urine visible fraction instance segmentation method

• Main Authors: ZENG QIANTAO, XU LIWEI, TU SHUQIN, YU HUIMING
• Format: Patent
• Language: Chinese; English
• Published: 19.09.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention relates to the technical field of image processing, and discloses a urine visible component instance segmentation method, which comprises the following steps: constructing a urine visible component annotation image data set, and dividing the urine visible component annotation image data set into a training set and a test set; taking a Mask Scoring R-CNN model as a basic model, and training an instance segmentation model fused with an SE attention mechanism by using the training set; and inputting urine visible component labeling data in the test set into the trained instance segmentation model for instance segmentation to obtain a urine visible component segmentation result. According to the method, on the basis of the Mask Scoring R-CNN model, the SE attention mechanism is fused, instance segmentation is carried out on the urine visible component image, the problems of urine visible component leak detection and urine visible component segmentation roughness caused by cell adhesion, overlapping