Driven by climate change concerns, wind energy has garnered significant attention, with the installed capacity of wind turbines rapidly increasing over recent decades. However, wind turbines located in harsh environments such as plateaus, mountainous regions, and offshore areas face complex operating conditions, leading to increased susceptibility to faults and wear, and resulting in high maintenance costs. This paper proposes a novel fault diagnosis method for wind turbines, leveraging multi-source vibration signal imaging feature fusion and a Transformer network architecture. The proposed method fuses multi-channel vibration signals to enrich fault information and leverages imaging features with a Transformer network to enhance fault feature extraction, achieving precise fault diagnosis for wind turbines. Experimental analysis of real-world wind turbine faults demonstrate that the proposed method achieves high diagnostic accuracy. Comparisons with other models in literature further indicate that the method offers superior diagnostic accuracy under real operating conditions.