This study evaluates different machine learning algorithms in detecting and identifying drones using radar data from a 60 GHz millimeter-wave sensor. These signals were collected from a bionic bird and two drones, namely DJI Mavic and DJI Phantom 3 Pro, which were represented in complex form to preserve amplitude and phase information. The first benchmarks used four algorithms, namely long short-term memory (LSTM), gated recurrent unit (GRU), one-dimensional convolutional neural network (Conv1D), and Transformer, and they were benchmarked for robustness under noisy conditions, including artificial noise types like white noise, Pareto noise, impulsive noise, and multipath interference. As expected, Transformer outperformed other algorithms in terms of accuracy, even on noisy data; however, in certain noise contexts, particularly Pareto noise, it showed weaknesses. For this purpose, we propose Multimodal Transformer, which incorporates more statistical features—skewness and kurtosis—in addition to amplitude and phase data. This resulted in a improvement in detection accuracy, even under difficult noise conditions. Our results demonstrate the importance of noise in processing radar signals and the benefits afforded by a multimodal presentation of data in detecting unmanned aerial vehicle and birds. This study sets up a benchmark for state-of-the-art machine learning methodologies for radar-based detection systems, providing valuable insight into methods of increasing the robustness of algorithms to environmental noise.

The newly proposed multimodal transformer architecture offers a new paradigm for UAV detection and aerial object recognition. It introduces an innovative way of feeding multiple data streams, such as audio, infrared video, RGB video, and radar, into the architecture for processing, using independent modalities. The unique features of each modality are attached and processed together in the architecture, where the features are then exposed to the multimodal transformer for classification. Thus, all complementary information can be pooled within the integration framework to allow the model discrimination of any drone target under outdoor conditions from other aerial objects such as birds, helicopters, and airplanes. These methodologies are expected to outperform traditional single-modality systems by improving detection accuracy through class balancing and addressing modality-specific limitations. The proposed model has been further tested through various experiments to evaluate its robustness under conditions such as missing entries, corrupted data, and synthetic inputs. The results suggest that it has strong potential to serve as a benchmark in UAV detection. Thus, this work takes part of an emerging body of sensor fusion and deep learning-related research, demonstrating the potential of multimodal data in real-world detection problems.

This work investigates the performance of different machine learning algorithms in the detection and identification of drones based on radar data from a 60 GHz millimeter-wave sensor. These signals are collected from a bionic bird and two drones, namely DJI Mavic and DJI Phantom 3 Pro, which were represented in complex form to hold amplitude and phase information. First benchmarks used four algorithms, namely LSTM, GRU, CNN, and Transformer, evaluated on noisy data. The main emphasis is on the assessment of algorithm robustness under noisy conditions, including artificial noise types like white noise, Pareto noise, impulsive noise, and multi-path interference. As expected, the Transformer outperformed other algorithms in accuracy, even on noisy data. However, certain noise contexts, especially Pareto noise, showed weaknesses in the effectiveness of the Transformer. For this purpose, we propose a Multimodal Transformer that incorporates more statistical features-skewness and kurtosis-in addition to amplitude and phase data. This resulted in a great improvement in the accuracy of detection, even under difficult noise conditions. Our results demonstrate the importance of noise in processing radar signals and the benefits afforded by a multimodal presentation of data in the problem of detecting UAVs and birds. The presented work sets up a benchmark for state-of-the-art machine learning methodologies for radar-based detection systems, providing valuable insight into methods of increasing the robustness of algorithms to environmental noise.