Practical Study of Recurrent Neural Networks for Efficient Real-Time Drone Sound Detection: A Review

The detection and classification of engine-based moving objects in restricted scenes from acoustic signals allow better Unmanned Aerial System (UAS)-specific intelligent systems audio-based surveillance systems. Recurrent Neural Networks (RNNs) provide wide coverage the field analysis due to their effectiveness widespread practical applications. In this work, we propose study SimpleRNN, LSTM, BiLSTM, GRU recurrent network models for real-time UAV sound recognition based on Mel-spectrogram using Kapre layers. main goal work is types RNN networks a sense reliable drone system. According results an experimental study, (Gated Units) model demonstrated higher prediction ability than other architectures detecting differences state signals. That is, RNNs gave CNNs loaded unloaded audio states various models, while showed about 98% accuracy determining load 99% background noise, which consisted more data.