The identification and characterization of glomeruli are crucial steps in the study of renal pathology. This paper presents a glomerulus segmentation method based on Convolutional Neural Networks (CNNs). Utilizing Whole Slide Images (WSI) and the ResNet101-UPerNet model, we achieve precise pixel-level segmentation of glomeruli. Through patch-level diseased glomeruli segmentation, we found that the model performs poorly in edge prediction. To address this problem, a center aggregation strategy is employed to integrate patch-level results into WSI-level outcomes. This method demonstrates exceptional performance across various mice PAS-stained models, achieving a Dice coefficient of 93.36 \(\%\) and an F1-score of 89.99 \(\%\) , ranking second and third in the MICCAI2024 KPIs2024 Task2 instance segmentation and detection competition. Results indicate that this approach offers good accuracy and versatility in WSI-level segmentation, establishing it as an effective tool for pathological image segmentation. Code is available at https://github.com/ZhiJianLife/KPIs2024_ZhiJianLife_task2 .