Detection and System Implementation of Abnormal Audio Events in Urban Environments Based on Convolutional Neural Networks

With rapid urbanization, detecting and recognizing abnormal audio events in urban environments is increasingly critical. This study systematically investigates methods for analyzing and processing such events and designs a detection system. Using the SONYC-UST-V2 dataset and Mel spectrogram features, models like DeepConvNet and MobileNetV2 were utilized for recognition. Results show these models perform well in both fine-grained and coarse-grained evaluations, exhibiting strong stability and generalization in multi-label classification tasks. The designed system can monitor, identify, and respond to abnormal audio events in real-time, enhancing data quality and usability.