Abstract:

To address the problems of high cost, asynchronous signals, and high false and missed alarm rates in complex scenarios of traditional single detection systems when fire smoke and methane leakage coexist in long-distance spaces, a linear-beam smoke-methane composite detection algorithm based on dual-wavelength characteristics was proposed. A long optical path spectral measurement platform with a total length of 120 m was constructed, the spectral characteristics of smoke attenuation and methane absorption were systematically analyzed, and a dual-wavelength detection architecture at 850 nm and 1 653.7 nm was designed. The algorithm adopted a three-branch parallel structure: the smoke detection branch learned the temporal attenuation characteristics of dual-wavelength signals through dual input channels. The methane detection branch extracted the morphological features of absorption peaks combined with an attention mechanism, and then captured dynamic signal changes via a GRU. The merging branch realized comprehensive judgment of three types of scenarios through feature fusion and temporal accumulation logic. Comparative experiments with six traditional fire and gas detection algorithms show that the proposed algorithm achieves an accuracy of 95.0%, false alarm rate of 2.3%, missed alarm rate of 1.7%, response time of 6.9 s and inference time of 0.28 s, with overall performance significantly superior to traditional methods.

Key words: dual-wavelength feature, linear-beam, smoke detection, methane detection, gated recurrent unit (GRU), composite detection