The application discloses an
engineering safety risk perception method based on
deep learning, which comprises the following steps: step one, collecting
engineering safety data of a construction site; step two, performing data preprocessing to generate a standard multimodal sample set; step three, generating a risk
perception tensor sequence through the improved FusionNet network from the standard multimodal sample set; step four, constructing an
engineering risk
perception vector sequence based on the risk perception
tensor sequence; step five, inputting the engineering risk perception vector sequence and the risk perception
tensor sequence into a risk reasoning model to generate a construction site risk perception result; step six, collecting safety event feedback information of the construction site, and updating the improved FusionNet network and the risk reasoning model; and step seven, redeploying the updated improved FusionNet network and the risk reasoning model. The application improves the perception ability of engineering safety risks through the improved FusionNet network.