Metal surface defect identification is the most critical task in quality control, where high accuracy and efficiency must be preserved so that neither the company nor the client suffers any difficulties. A deep learning technique, Convolutional Neural Network (CNN) can be applied to identify various defects in steel surfaces. Hence, to enhance CNN’s performance, the Gray-Level Co-occurrence Matrix (GLCM) algorithm is applied to extract texture features. These handcrafted texture features are combined with spatial features learned by CNN in a hybrid framework. The outputs from both modules are fused and passed through a unified deep neural network to leverage complementary information. The proposed model is evaluated on the benchmark NEU Metal Surface Defects Dataset and compared with a baseline CNN model to demonstrate the improvements in classification accuracy and reliability.