Kidney pathology image segmentation aims to accurately extract tissues and lesions but is challenging due to image complexity and tissue similarity, especially for non-experts. The high resolution of WSI images further complicates processing, requiring significant computational resources. To address this, we propose an Adaptive Crop method that utilizes optical magnification to maintain scale consistency between cropped patches and training samples. A boundary removal mechanism prevents edge effects when glomeruli are near the patch edges. Overlapping crops and stitching predicted results enhance the accuracy and robustness of WSI image segmentation. Our code and trained models are publicly available at https://github.com/jasminethurder/KPI2024 .