Vibration-based structural health monitoring methods have been widely applied in the field of damage identification. This paper proposes an intelligent diagnostic approach that integrates a multi-scale convolutional neural network with a multi-head attention mechanism (MSCNN-MHA) for the structural health monitoring of jacket-type offshore platforms. Through numerical simulations, acceleration response signals of three-pile and four-pile jacket platforms under random wave excitation are analyzed. Damage localization studies are conducted under simulated crack and pitting corrosion cases. Unlike previous studies that often idealize damage by weakening structural parameters or removing components, this study focuses on small-scale damage forms to better reflect real engineering conditions. To verify the noise resistance of the proposed method, noise is added to the original signals for further testing. Finally, experiments are conducted on the basic structure of the jacket-type offshore platform, simulating small-scale crack and pitting damage under sinusoidal and pulse excitation, to further evaluate the applicability of the method. Compared to previous CNN and MSCNN-based approaches, the results of this study demonstrate that the MSCNN-MHA method achieves higher accuracy in identifying and locating minor damage in jacket-type offshore platforms.