Image-Free Infrared Detection Paradigm for Industrial Gas Leaks

Industrial gas leaks create hazardous situations that can lead to fires and explosions, making their detection and monitoring crucial. The optical gas imaging (OGI) has proven to be an attractive way to ensure safe working conditions. However, existing implementations for OGI schemes rely on pixelated sensors for image capture that produce abundant redundant information. Low computational overhead and real-time gas leak detection remain challenging. In this article, we propose an image-free gas leak detection approach that enables the direct extraction of leakage features from 1-D signals. Aimed at image-free measurement, the proposed method optically samples the partial Fourier spectrum by combining a single-scanline sensor with a Fourier optics architecture. Moreover, to effectively mine sequence features, the wavelet-fused convolution-transformer network architecture is proposed to deeply fuse a convolutional neural network, which has the ability of local sensing, with a transformer, which has the advantage of global modeling, in the feature space by using the wavelet transform. Extensive validation experiments of the proposed method are carried out on a self-constructed image-free gas leakage dataset and a very domain-representative industrial scene, IOD-Video. The results show that the proposed image-free detection paradigm exhibits significant advantages, achieving a detection accuracy of 96.4% in an inference time of 3.28 ms.