<p>Human pose estimation is a fundamental task in computer vision, with applications in sports analysis and medical rehabilitation. However, high-performance models are often computationally expensive, while lightweight models usually suffer from accuracy degradation. To address these issues, this paper proposes Slim-HRNet, a lightweight human pose estimation framework based on high-resolution networks (HRNet). First, we construct an efficient variant by pruning the fourth stage of the HRNet, while retaining the feature representation capability of its multi-resolution branch. To further enhance computational efficiency, we design a lightweight enhanced basicblock (LE-Basicblock) by combining depthwise separable convolution (DW) and efficient channel attention. In addition, multi-scale feature enhancement fusion module is designed to strengthen cross-scale feature interaction and improve the context representation ability of pose structure. DySample is adopted for efficient upsampling, and a heatmap decoding refinement strategy is proposed to enhance keypoint localization accuracy. Experimental results on the COCO2017 and MPII datasets demonstrate that Slim-HRNet achieves 90.1% PCKh on MPII with only 5.74M parameters and 4.46 GFLOPs, and attains 72.6% AP on COCO with 3.35 GFLOPs. This represents an approximately 80% reduction in parameters and a significant increase in inference speed to 38.51 FPS. These results show that Slim-HRNet improves efficiency while maintaining competitive accuracy.</p>