Multi-scale Feature Fusion Assisted: Fresh Air Pipeline Leakage Detection Using an Improved Faster-RCNN

To address the issue of the accuracy degradation caused by high-noise environments in fresh air pipeline leakage detection, we propose an improved Faster-RCNN model that innovatively integrates the advantages of channel-spatial attention and feature pyramid network (FPN). Concretely, channel–spatial attention is used to adaptively suppress non-target noise regions within spectrograms, thereby mitigating interference during leakage detection. The FPN is utilized to fuse multi-scale time-frequency features to further enhance the performance of leakage detection in high-noise environments. The proposed model is evaluated on our dataset that includes different noises, including turbulence, rock impacts, and wind. Experimental results show that under low signal-to-noise-ratio (5 dB) scenarios, the proposed model achieves a detection accuracy of 97.2%, outperforming the baseline Faster-RCNN by 6.3 %. Moreover, compared to the baseline model, the false-alarm rate of the proposed model is reduced to 2.8%, demonstrating the robustness and practicality of the proposed model in high-noise environments.