Abstract: Because of existing texture patterns such as straight, parabolic and chaotic, this paper proposes a fast and accurate classification method. First, 15 features from wavelet and 16 features from curvelet transform were extracted. Then, a genetic network was designed, whose inputs represent 31 features and outputs represent 3 texture patterns. After genetic operation, 14 features were optimized. Finally, texture classification was constructed based on BP network using the optimized features. Tests were conducted for 300 samples, categorized into three types. The average classification rates of the wavelet, the curvelet and the fusion methods were 86.5%, 89.3% and 90.9%, respectively. The classification times were 0.025, 0.563 and 0.216 seconds. Experimental results showed that wavelet transform had good classification for straight textures, but struggles with complex textures lack of direction. Curvelet transform can be used to express the complex texture of wood, but the computational time for its features is long. The genetic fusion method combines the fast classification of wavelet and the high accuracy of curvelet by extracting the effective features for classification.