A two-dimensional principal component analysis algorithm based on the L1-norm and the F-norm is proposed for the detection of concrete cracks in shield tunnel segments. In light of the fact that the resolution of the issue of outliers is of greater consequence in the context of practical engineering, the L1-norm metric is employed to attenuate the sensitivity of feature extraction algorithms to outliers. Concurrently, the F-norm metric is employed to diminish the reconstruction error of the algorithm, augment its reconfigurability, and facilitate more accurate crack labelling. Subsequently, the efficacy of the proposed algorithm was evaluated through tests on concrete crack images. The results demonstrated that the algorithm exhibited commendable recognition and labeling capabilities, achieving a recognition rate of up to 90.42%. Furthermore, an analysis of concrete crack detection under varied experimental conditions was conducted, with the results indicating the efficacy of the proposed algorithm in mitigating noise. Finally, the algorithm is applied in the field of face recognition, and the experimental results show that the proposed algorithm still has strong robustness and practical applicability. In summary, this evidence substantiates the assertion that the strategy of employing two-dimensional principal component analysis-related algorithms is a promising avenue for concrete crack detection. Moreover, it paves the way for future research endeavors aimed at further refining and enhancing these algorithms.