Study of Urban Environmental Noise Source Identification Technology Using Deep Learning

Urban noise pollution has become a serious environmental issue affecting human health and quality of life. To strengthen the precise and efficient management and control of environmental noise pollution, an efficient and accurate environmental noise source identification technology is urgently needed. This study investigates the environmental noise sources that significantly affect residents in urban areas, represented by Shanghai, and collects a typical urban environmental noise dataset. Further, based on the collected dataset, a Convolutional Recurrent Neural Network (CRNN) model for the automatic identification of urban environmental noise sources is designed. The model integrates a 3-layer Convolutional Neural Network (CNN) module and a 2-layer Gated Recurrent Unit (GRU) module, which not only has the ability to extract high-dimensional spectral features but also captures temporal information of features. The training results show that the model's average identification accuracy (Accuracy, Acc) reaches 93%, with Acc values for all nine classes exceeding 90%, and the lowest Acc value reaching 86.6%. This verifies the effectiveness and feasibility of the CRNN model-based urban environmental noise source identification technology. Moreover, the model's identification Acc is significantly better than that of CNN (with an Acc value below 30%) and GRU (with an Acc value of about 78%), further verifying that the integrated CRNN method is more suitable for processing environmental noise source identification tasks.