The Czochralski silicon single-crystal growth process exhibits strong coupling, dynamic behavior, and pronounced nonlinearity. In production, weak anomalies and slight process parameter fluctuations are often difficult to detect in time, yet they may gradually accumulate and affect the stability of crystal diameter and product quality. To address this issue, a process monitoring method integrating feature enhancement and discriminant optimization, termed hybrid cosine-neighborhood preserving embedding with Fisher discriminant canonical variate analysis(HC-NPE-FDCVA), is proposed. The method follows a progressive modeling framework consisting of feature extraction, dynamic modeling, and state discrimination. First, considering the high-dimensional, nonlinear, and multimodal characteristics of the process data, the hybrid cosine-neighborhood preserving embedding(HC-NPE) algorithm is introduced for nonlinear feature extraction and dimensionality reduction. By combining cosine distance with Euclidean neighborhood structure information, the algorithm preserves the local manifold structure of the original data while enhancing the representation of weak abnormal patterns. Subsequently, Fisher discriminant canonical variate analysis(FDCVA) is applied in the reduced feature space. By constructing a discriminative dynamic state subspace based on the Fisher criterion, differences between normal operating states and abnormal conditions are effectively separated, thereby improving the distinguishability of anomaly-related features in the state space. On this basis, an online process monitoring model is established. Monitoring indices are constructed using the T2 and Q statistics, enabling real-time evaluation and anomaly identification for the silicon single-crystal growth process. Experimental results demonstrate that the proposed method maintains high anomaly detection sensitivity while effectively reducing the false alarm rate. It can more effectively characterize the evolution of abnormal behaviors in complex nonlinear industrial processes, providing a feasible monitoring approach for the stable operation and quality control of Czochralski silicon single-crystal growth processes.