Deep-Learning-Based Gas Leak Source Localization From Sparse Sensor Data

Detection and Estimation of Gas Sources With Arbitrary Locations Based on Poisson's Equation

“…Bayesian inference on a linear model with such a hierarchical prior is commonly realized via two types of techniques: MAP estimation of w (Type I estimation 3 ) and MAP estimation of α (Type II estimation). Type II estimation commonly demonstrates improved performance in contrast to Type I estimation.…”

“…Accurate concentration mapping of the airbourne substances can be realized through the process of "learning" from data, employing data-driven methodologies. Various approaches have been proposed to address this objective, including support vector machines for concentration map-ping [1], kernel methods [2], and deep-learning techniques, albeit predominantly applied within the visual domain [3] [4]. One notable limitation of these methodologies lies in their reliance on training data, particularly evident in the case of deep-learning methods.…”

“…In[42] the authors extend the framework by generalizing p(w l |α l ) to be the pdf of a power exponential distribution, which makes the hierarchical prior a power exponential scale mixture distribution 3. Note that many traditional "non-Bayesian" methods for learning sparse representations such as basis pursuit de-noising or re-weighted p-norm regressions[43],[44],[45] can be interpreted as Type I estimation within the above Bayesian framework (see[42] for more details on this link).…”