Long-Distance Field Demonstration of Imaging-Free Drone Identification in Intracity Environments

Abstract

Detecting small objects, such as drones, over long distances presents a significant challenge with broad implications for security, surveillance, environmental monitoring, and autonomous systems. Traditional imaging-based methods rely on high-resolution image acquisition, but are often constrained by range, power consumption, and cost. In contrast, data-driven single-photon-single-pixel light detection and ranging (\text{D\textsuperscript{2}SP\textsuperscript{2}-LiDAR}) provides an imaging-free alternative, directly enabling target identification while reducing system complexity and cost. However, its detection range has been limited to a few hundred meters. Here, we introduce a novel integration of residual neural networks (ResNet) with \text{D\textsuperscript{2}SP\textsuperscript{2}-LiDAR}, incorporating a refined observation model to extend the detection range to 5~\si{\kilo\meter} in an intracity environment while enabling high-accuracy identification of drone poses and types. Experimental results demonstrate that our approach not only outperforms conventional imaging-based recognition systems, but also achieves 94.93\% pose identification accuracy and 97.99\% type classification accuracy, even under weak signal conditions with long distances and low signal-to-noise ratios (SNRs). These findings highlight the potential of imaging-free methods for robust long-range detection of small targets in real-world scenarios.

Figures (9)

• Figure 1: Conclusion of single-photon LiDAR target identification methods.
• Figure 2: a Demonstration of D2SP2-LiDAR. b Target identification using single-pixel imaging. c Raster scanning imaging using galvanometer. The transceiver is composed of a pulsed laser and a single-pixel SPD. d Imaging using array SPD. e Comparison of distance, cost, and power consumption of different single-photon target detection methods. Our work shows low cost and low-power consumption while maintaining high accuracy in long-range scenarios.
• Figure 3: a Overall architecture of 1D-ResNet. The network is composed of a feature extractor and a linear classifier. Temporal histograms are fed to the feature extractor to generate features, then the linear classifier maps the features to the number of poses or types. b Detailed structure of the residual block, which is composed of two 1D convolutional blocks and a shortcut adding the input with the convolution output.
• Figure 4: The generated poses of the drone model in simulation, shown as depth maps.
• Figure 5: Average accuracy of different neural networks for 18 poses of drone model (see Fig. \ref{['fig4']}). a$\sim$d are results for different reflected photon number, $N_s=5000, 500, 250, 125$. e is the average test accuracy comparison, which is averaged under all SNRs and $N_s$.