LAUSR

For the Visual Internet of Things (VIoT), heterogeneous sensors such as visual cameras and acoustic microphones can be installed in devices to perform cognitive sensing and collect multiview data. Nonetheless, not every device has the same set of sensors due to deployment costs or sensor malfunctioning. This subsequently causes incomplete heterogeneous data, which means that data from various sensors may not be intact. For example, visual data are present, whereas audio signals are unavailable. Such phenomena may become severe when a large scale of VIoT devices are involved during crowdsensing, not to mention that missing values might occur in the collected data. In view of such, this study proposes solutions based on half quadratic (HQ) optimization to conquer the above problems in the VIoT. In this article, challenges involving incomplete heterogeneous multiview (IHM) data are investigated: IHM data reconstruction, IHM feature extraction, and IHM data recognition. Moreover, the corresponding solutions to the above challenges - HQ matrix completion, HQ supervised discrete hashing, and HQ graph neural networks - are also introduced in this article. Experiments on various HQ functions were carried out to examine their effectiveness, and the experimental results verify the proposed solutions.