Scalable and low-power edge architecture with Wi-Fi HaLow and on-device spectrograms generation for flexible urban bioacoustics monitoring

• Real-time bird classification using edge AI in urban smart cities. • Low-power IoT nodes transmit only relevant spectrogram snippets. • Automatic Mel-spectrograms enable continuous CNN training. • Dual Wi-Fi/Wi-Fi HaLow connectivity supports scalable deployments. • Detects native and invasive species (e.g., Psittacula krameri). Urban biodiversity monitoring in smart cities requires scalable and efficient computing architectures capable of handling real-time, distributed sensing tasks. This paper proposes a low-power edge computing and Internet of Things (IoT) framework that enables on-device acoustic detection and classification of bird species, serving as bioindicators of ecosystem health. The architecture leverages lightweight convolutional neural networks (CNNs) deployed on energy-efficient sensor nodes, significantly reducing communication overhead by transmitting only detection events and compact spectrogram data. A key contribution is the automatic generation of Mel-spectrograms at the edge, which supports the continuous creation of training datasets and iterative neural network refinement without manual preprocessing. The proposed system incorporates dual Wi-Fi and Wi-Fi HaLow connectivity, providing adaptable long-range, low-power communication for heterogeneous urban environments. Field experiments validate the framework’s scalability and effectiveness, demonstrating robust detection of both native and invasive species. By combining distributed intelligence, resource-aware computation, and flexible networking, the system offers a practical edge–IoT solution for large-scale, real-time environmental monitoring in smart city contexts.