Functional canonical correlation analysis is a key method in multivariate statistics for identifying optimal linear correlations between two functional datasets. However, some functions within these datasets may exhibit anomalies such as sudden changes or fluctuations that deviate from the overall trend, resulting to inaccurate results. To address this, we propose an improved method: Sparse functional canonical correlation analysis based on the L_2,1-norm. This approach reduces outliers by optimizing the selection of orthogonal basis functions, thereby enhancing the accuracy and reliability of the analysis. Numerical experiments show that the L_2,1-norm-based method significantly outperforms traditional methods.