Enhanced Multi-Modal Gas Leakage Detection with NSMOTE: A Novel Over-sampling Approach

Aref Azizian, Behnam Yousefimehr, Mehdi Ghatee

Abstract: Urbanization and the emergence of smart cities have led to increased industrial activities, raising concerns about safety and environmental hazards. Detecting and monitoring gas leaks accurately is crucial due to the significant risks they pose to public health and the environment. In this study, we propose a novel approach that combines thermal imaging with sensor data to enhance gas detection accuracy in urban and industrial settings. By employing advanced algorithms and analyzing extensive datasets, our methodology achieves an impressive accuracy rate of 99.32%, setting a new standard in gas leak detection. Our innovation lies in the utilization of a unique oversampling technique called Normal Synthetic Minority Oversampling Technique (NSMOTE), coupled with the simultaneous integration of thermal imaging and sensor data within a multimodal framework. This breakthrough not only underscores the potential of integrating diverse data sources for environmental monitoring but also represents a significant stride in mitigating the risks associated with industrial gas emissions.