Deep learning-based image steganography for visual data cybersecurity in construction management

The construction industry is increasingly digital and dependent on extensive use of information technologies. However, data exchange in a digital environment makes construction data more vulnerable to cyber risks. For instance, construction videos contain various site information (such as worker privacy, innovative techniques, and infrastructures status), the loss of which may cause financial and safety issues. To ensure the cybersecurity of visual data in construction, this research proposes a deep learning-based image steganography method, which can cover the secret image with an irrelevant image by using a hidden neural network and retrieve the secret image with a reveal neural network. In experiments, a dataset containing 7,000 construction images was used for validating the feasibility of the proposed method. Three evaluation metrics were used to test the performance of proposed method in visual information hiding and recovery. Specifically, the proposed method achieved a peak signal-to-noise ratio of 36.58, a structural similarity index of 97.29%, and a visual information fidelity of 82.57% on average. The test results demonstrate the reliable performance of the proposed method in protecting construction visual data. This research provides a novel way to ensure the cybersecurity of visual data in construction, other than simple password encryptions.