To accurately identify the safety hazards in complex hydropower engineering construction scenes in real time, an automatic identification method of safety hazards in hydropower engineering construction based on enhanced feature representation was proposed. The feature extraction network was constructed, and the squeeze and excitation module was embedded to adaptively enhance the feature expression, enhance the identification effect of the image features of the safety hazards, and reduce the influence of background noise. The feature enhancement network was constructed, and the group-wise separable convolution module and the vision-oriented ghost spatial cross-stage partial convolution module were introduced to alleviate the loss of low-level detail information, enhance the feature fusion ability, and improve the identification accuracy of multi-scale safety hazards. The results of the engineering case verification show that the model can overcome the interference of complex scenes by strengthening the feature expression, and the average accuracy of construction safety hazard identification is up to 86.8%, which is better than the existing hydropower engineering construction safety hazard identification model and provides technical support for the intelligent and refined management of hydropower engineering construction safety hazards.