Wind energy is recognized as a renewable source due to its reliability and low cost. However, under cold and humid conditions, blade icing poses a serious hazard to the performance and durability of wind turbines. Millimeter-wave radar-based inspection techniques have gained attention for their ability to penetrate non-polarized materials and provide surface conditions and in-depth information independently of light and weather conditions. A real-time detection method for wind turbine blade icing using 77 GHz millimeter-wave radar is proposed. Mel-frequency cepstral coefficients (MFCCs) are extracted from mixed-frequency time-domain signals and fused with one-dimensional convolutional neural network (1D-CNN) for classification and identification of blade icing types. The effectiveness of the proposed method is verified through experiments with varying distances and directions. Four icing types and different thicknesses are accurately recognized, achieving a recognition rate of 94%. Thin ice coverage on wind turbine blades can be recognized and warned against.