Abstract: According to the characteristics of construction site management, this paper introduces real-time video analysis of the application of cnooc construction scene hidden dangers and model optimization. To solve the problem of Marine engineering field workers without wearing helmet, not wearing clothes, right into the forbidden area, etc., avoid because of irregularities or in violation of regulations, unnecessary casualties or damage, based on a set of cnooc video identification of the alarm system on the construction site, design a set of comprehensive utilization of helmet color, outline, characteristics of clothing and multi-layer neural network classification statistical model is set up, whether to wear safety helmet and correctly wear clothing, whether into no-go area for testing. Thus, to a certain extent, put an end to safety hazards.