The segmentation of pole pieces in CT images is a crucial step in utilizing industrial computed tomography （CT） to detect the pole pieces of laminated cells. However， the complex structure and high aspect ratio of laminated cells pose challenges for existing segmentation methods， which struggle to meet the speed and accuracy demands of large-scale production. To address this， this paper introduces a novel CT image segmentation network based on a strip attention mechanism. The proposed network comprises a lightweight backbone （MobilenetV2）， a Strip Atrous Spatial Pyramid Pooling （S-ASPP） module， and a decoder module. MobilenetV2 reduces network parameters and enhances segmentation speed， while the S-ASPP module employs strip pooling to retain strip feature information， effectively mitigating under-segmentation issues. The strip attention mechanism in the decoder focuses on edge and detail information， improving edge sharpness and detail clarity. Experimental results demonstrate that the proposed network achieves an average pixel classification accuracy （mAcc） of 98.62%， an average intersection over union （mIoU） of 89.97%， a parameter count of 5.814M， and a segmentation speed of 56.94 frame/s. Compared to DANet， DeepLabV3+， U-Net， HRNet， and Segnext， the proposed method achieves the highest segmentation accuracy and outperforms DANet， DeepLabV3+， U-Net， and HRNet in speed， with slightly lower speed than Segnext. Considering mIoU， mAcc， FPS， and parameter count comprehensively， the proposed network significantly enhances segmentation precision and efficiency while maintaining low computational cost， outperforming mainstream segmentation networks.