YAMNet-based transfer learning for compact noise classification in urban and wireless systems

Urban development and construction generate significant noise, which must be accurately monitored and classified to effectively mitigate its impact and ensure regulatory compliance. However, traditional noise recognition methods tend to perform poorly in complex, dynamic environments because they rely on large labeled datasets and have limited adaptability. To overcome these challenges, this study proposes a novel noise source classification framework YAMNet-Trans, a novel noise classification framework that synergizes transfer learning with acoustic and communication signal processing techniques. By fine-tuning a pre-trained YAMNet model and integrating Mel-frequency cepstral coefficients (MFCC) and short-time Fourier transform (STFT)—techniques widely used in communication signal processing for tasks such as speech enhancement and channel equalization—the proposed method achieves robust feature extraction and high classification accuracy while using minimal labeled data. Experimental results show that the method attains a classification accuracy of 94.21% with limited training samples, outperforming benchmark models including ResNet-50, VGG-16, AST, BEATs, and the baseline YAMNet. Furthermore, to validate the applicability of the proposed framework within the communications field, MFCC and STFT were applied for denoising on the RML2016.10a modulation dataset, and the model was subsequently used on the denoised dataset to obtain highly accurate classification results. This work not only advances urban noise management but also bridges acoustic signal processing with communication technologies, showcasing potential applications in real-time anomaly detection for IoT networks in smart cities and within 5G infrastructure. By combining high accuracy, computational efficiency, and adaptability to resource-constrained environments, YAMNet-Trans offers a versatile solution for both environmental noise management and next-generation communication systems.