Abstract: Surface roughness is recognized as a key indicator of machined surface quality and service performance. To overcome the limitations of noncontact roughness measurement, including weak feature representation, low accuracy, and poor real-time performance, an online detection model based on multi-parameter feature fusion is proposed. From surface images, grayscale, statistical, and texture features are derived, and key parameters are determined through correlation analysis to construct a representative feature set. The association between the extracted features and surface roughness is formulated within a support vector regression (SVR) model, while its parameters are refined through adaptive optimization using the gray wolf optimization (GWO) algorithm to enhance prediction accuracy and robustness. Experimental evaluation indicates that mean absolute errors of 0.0279 μm and 0.0409 μm were obtained under dry and mist cooling conditions, respectively. A detection speed exceeding 46.45 FPS was recorded, confirming that high accuracy and real-time performance can be simultaneously realized for intelligent surface quality monitoring.