Journal of Multimedia Processing and Technologies Qianwei Zhang, Lirong Yu 17 1 2026 https://doi.org/10.6025/jmpt/2026/17/1/1-10 https://www.dline.info/jmpt/fulltext/v17n1/jmptv17n1_1.pdf The work proposes an improved method for assessing the motion stability of badminton athletes using advanced image segmentation techniques. It highlights the growing importance of sports stability in performance enhancement and scientific training. Traditional video segmentation methods are noted as insufficient for accurate athlete segmentation, prompting the integration of deep learning with fuzzy neural networks. The authors present a hybrid algorithm combining the computational efficiency of neural networks with the precision of fuzzy logic to better handle illumination changes and motion dynamics in video frames. Experimental results demonstrate that their genetic fuzzy neural network algorithm achieves significantly lower error rates 5.01% and 2.52% compared to traditional and standard fuzzy neural approaches, which often exceed 15% error. The method shows high accuracy in segmenting moving targets, such as badminton players, enabling clear foreground background differentiation and reliable feature extraction. Performance tests confirm improvements in both processing speed and segmentation quality, especially for complex motion scenarios. Despite these advances, the study acknowledges a limitation: it does not address the methodology for movement step analysis. Overall, the research contributes a feasible, accurate, and efficient image segmentation solution tailored for sports stability analysis, leveraging adaptive deep learning to support real world athletic performance evaluation and decision making.