Pith. sign in

REVIEW 4 major objections 6 minor 49 references

Fusing ISAC micro-Doppler spectra with PTZ camera images distinguishes UAV and bird subtypes at about 98% average accuracy across a wide SNR range.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-30 12:24 UTC pith:RT3MQAW6

load-bearing objection Solid first ISAC–vision fusion stack for 10-way UAV/bird recognition, but the headline accuracy is measured with oracle PTZ pose, not the ISAC estimates the paper advertises. the 4 major comments →

arxiv 2607.23789 v1 pith:RT3MQAW6 submitted 2026-07-26 eess.SP

ISAC and Vision Fusion for Fine-Grained Low-Altitude Target Recognition

Zhonghua Chu , Hongliang Luo , Boxuan Sun , Shengjie Quan , Chuanbin Zhao , Feifei Gao This is my paper
classification eess.SP
keywords integrated sensing and communicationsISAC and vision fusionmulti-modal ISACfine-grained low-altitude target recognitionmicro-DopplerPTZ cameraMobileViTlow-altitude economy
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Low-altitude airspace is filling with drones, eVTOLs, and birds that look similar from afar but pose different risks. This paper argues that a base station already doing integrated sensing and communications can both locate a tiny target and cue a pan-tilt-zoom camera to photograph it from hundreds of meters away, then fuse the wireless micro-Doppler signature with the cropped visual patch to name the exact subtype. A conditional GAN cleans noisy Doppler images; YOLOv11 isolates the target in the photo; and a dual-stream MobileViT network merges the two modalities. On a new simulated joint dataset spanning ten classes, multiple ranges, urban and open scenes, and sunny/rainy/snowy weather, the fused system holds near 98% accuracy from 0 to 30 dB SNR while vision alone or radar alone lag far behind. If the approach transfers beyond simulation, existing 6G-style base stations plus ordinary PTZ cameras could form a fine-grained low-altitude monitoring layer without dedicated radar deployments.

Core claim

An ISAC base station that estimates target position, steers a co-located PTZ camera, extracts and cGAN-denoises the micro-Doppler spectrum, crops the visual region of interest with YOLOv11, and fuses both streams in a MobileViT dual-branch network can recognize ten fine-grained low-altitude classes (five UAV models and five bird species) at 97.67% average test accuracy across 0–30 dB SNR on the authors’ joint ISAC–vision dataset, substantially outperforming noisy-fusion, vision-only, and Doppler-only baselines.

What carries the argument

The dual-stream MobileViT fusion network: MobileViT-S on the cGAN-cleaned micro-Doppler image and MobileViT-XS on the YOLOv11-cropped RGB patch, with features concatenated after a projection layer and classified by an MLP. ISAC position estimates close the loop by setting the camera’s pan, tilt, and zoom so the tiny target is large enough in the frame to be useful.

Load-bearing premise

The dataset and evaluation assume clean line-of-sight channels and camera frames pointed with ground-truth target pose rather than the noisy position estimates the real closed loop would feed the camera.

What would settle it

Re-run the full pipeline on the same trajectories but drive the PTZ from the actual noisy ISAC position/velocity estimates (and include multipath) and check whether the fused accuracy still stays near 98% down to 0 dB SNR; a large drop would falsify the claimed closed-loop gain.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Future 6G base stations can double as the sensing backbone of a low-altitude monitoring network without separate dedicated radars.
  • Micro-Doppler cleaned by a cGAN remains informative enough at low SNR to rescue vision when weather or distance degrades the optical image.
  • Active PTZ control from wireless position estimates makes distant, pixel-small targets usable for fine-grained visual recognition.
  • The released-style JIVD construction (AirSim + Wireless InSite) supplies a reproducible multi-modal benchmark for ten UAV/bird subtypes under weather and range variation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If multipath and estimation error prove costly, a lightweight online recalibration that re-zooms after the first visual detection could restore much of the reported gain without new hardware.
  • The same ISAC-cued PTZ loop could extend to other small, high-threat classes (e.g., hobby rockets or balloons) once their micro-motion and appearance models are added to the dataset pipeline.
  • Binary UAV-versus-bird separation already looks linearly separable in the fused feature space; a cheap early rejector could gate the full ten-class head for edge deployment.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes an end-to-end ISAC–vision fusion framework for fine-grained low-altitude target recognition. An ISAC base station (26 GHz OFDM, hybrid and radar URAs) estimates target position/velocity via 2D-FFT plus MVDR refinement (§III.B); these estimates drive a PTZ camera's pan/tilt/zoom through the control laws in Eqs. (26)–(30). From the echo, an STFT micro-Doppler (mD) spectrum is extracted (§IV.A) and denoised by a cGAN (U-Net generator, PatchGAN discriminator, §IV.B); YOLOv11 crops the visual ROI (§V.A); a dual-stream MobileViT-S/XS network with feature projection and an MLP head fuses the two modalities for 10-class recognition (§V.B). A new synthetic dataset, JIVD, is built from AirSim trajectories replayed for image capture and Wireless InSite X3D ray tracing for channel parameters (§VI.A–B). Reported results: cGAN denoising improves MSE/PSNR/SSIM (Fig. 8) and mD-only accuracy (Fig. 9); YOLOv11 achieves max F1 = 0.995 at IoU 0.5 and mean IoU 0.849 (Figs. 10–11); the full fusion scheme averages 97.67% accuracy over SNR 0–30 dB versus 86.25% (noisy mD + vision), 74.34% (vision only), and 71.08% (denoised mD only), with a near-diagonal confusion matrix at 30 dB (Figs. 12–13) and well-separated UMAP clusters (Fig. 14).

Significance. If the results hold, this is a useful contribution: to my knowledge it is the first ISAC–vision fusion framework for fine-grained low-altitude recognition, and it ships several concrete assets — a complete, reproducible-in-principle pipeline from OFDM echo model to fused classifier; a new paired RF–vision synthetic dataset (JIVD) with 10 fine-grained classes, multiple ranges, scenarios, and weather; and a clean ablation structure (four schemes across seven SNR levels) that makes the fusion and denoising claims individually falsifiable. The reported fusion gains are large and consistent across SNR, and the mD signatures in Fig. 3 are physically plausible. The main caveats on significance are that all evidence is synthetic (AirSim + Wireless InSite with LoS-only channels), and — more importantly — the advertised closed-loop system (ISAC estimate → PTZ pointing → recognition) is not the system actually evaluated, since image acquisition uses simulator ground-truth pose. The paper therefore currently demonstrates separability under oracle acquisition rather than performance of the deployed loop.

major comments (4)
  1. [§VI.A vs. §III.C (Eqs. (26)–(30)); Fig. 12] The central claim concerns a closed-loop system in which ISAC position estimates (r̂, θ̂, φ̂) drive the PTZ camera. However, the evaluation pipeline records AirSim ground truth and then 'actively configure[s] the PTZ camera based on the target's current location' — i.e., every image in JIVD is acquired under oracle pose, not under the noisy estimator output. The zoom law (Eq. (30)) targets only L_cls = 40 pixels of target width at ranges up to 600 m, so small range/angular errors, estimator latency, or synchronization error can decenter the target, mis-set the focal length, or drop it from the FOV entirely. No estimation errors for (r̂, θ̂, φ̂) are ever reported, and they are never propagated through Eqs. (26)–(30). Consequently the 97.67% average accuracy and the fusion-over-vision margins in Fig. 12 may depend on perfectly centered and scaled images. This is fixable within the paper's
  2. [§III.C, Eq. (28)–(30)] The adaptive FOV law requires L_obj, the physical width of the target, as an input. But L_obj is class-dependent, and the target's subcategory is precisely what the system is trying to recognize; at acquisition time the target may not even be classified at the UAV/bird level. As written, the zoom control implicitly assumes knowledge of the target's size (or class) before recognition, which is circular for the claimed operational use. The authors should state what L_obj is used in practice (a fixed nominal value? a coarse estimate from radar cross-section or range-Doppler extent?), and quantify how mis-specification of L_obj propagates through Eq. (30) into L_cls and final accuracy.
  3. [§VI.A] The channel synthesis retains only the LoS path ('we focus exclusively on the LoS path and neglect NLoS'), yet the scenarios include 'dense urban building complexes,' where multipath is precisely the regime in which both mD extraction and MVDR-based position estimation degrade. Since the simulation framework (Wireless InSite X3D) already computes NLoS components, including at least a representative multipath setting — or explicitly scoping the claims to LoS-dominated geometries and removing the urban-complexity implication — is needed for the accuracy numbers to be interpretable.
  4. [§VI.B–E] The train/test protocol is under-specified in a way that bears on the validity of the headline numbers. The 42,000 noisy echoes are 7 SNR variants of 6,000 base sequences; the text says 80%/20% of 'sequences' are used for train/test, but it is not stated whether the split is performed before SNR augmentation (so no base trajectory appears in both splits at different SNRs) or after. Please clarify, and also report: training hyperparameters (optimizer, learning rate, epochs, batch size, the value of λ_loss in Eq. (46)), the number of test samples underlying each point in Fig. 12, and confidence intervals or per-sequence variance. A single split with no error bars is weak support for accuracy differences quoted to two decimal places.
minor comments (6)
  1. [§VI.E, Fig. 13] The confusion matrix is shown only at 30 dB, where performance is near-ceiling. A confusion matrix at a low SNR (e.g., 0–10 dB) would be more informative about which class confusions drive the degradation in Fig. 12.
  2. [§III.B] The performance of the 2D-FFT + MVDR position/velocity estimator is never quantified (no RMSE vs. SNR or range). Since Stage 1 feeds both the PTZ control and the receive beamforming in Eq. (31), at least a basic estimation-accuracy curve should be reported.
  3. [§V.B, Fig. 5] No baseline comparison to existing radar–vision fusion methods (e.g., [21]–[23]) and no complexity/latency analysis are provided. Given the claimed deployment motivation (edge monitoring), inference time and parameter counts for the dual MobileViT + cGAN + YOLOv11 stack would strengthen the paper.
  4. [§III.C, Eq. (27)] The quaternion conversion omits the roll angle γ entirely, although ω is defined with three components; with γ = 0 the standard ZYX conversion contains additional cross terms (e.g., the q_x and q_y entries involve both ψ and ϑ half-angle products in a specific order). Please state the rotation convention and verify the formula.
  5. [Throughout] Several typos and notation inconsistencies: 'adiust' (§III.C), 'singals' (Fig. 2), 'genetated' and 'alititude' (Fig. 7 caption), 'MobileVit' vs. 'MobileViT' (Fig. 5), 'UA V' spacing artifacts. Fig. 5's feature-dimension annotations render as 'ℝ????'. Please proofread.
  6. [§VI.B] The dataset size derivation (10×5×5×24 = 6,000) is hard to follow: it is unclear what the two factors of 5 represent (5 distances and 5 what — scenarios/weather/sequences?). Please make the per-class sequence count and the scenario/weather assignment explicit.

Circularity Check

0 steps flagged

Empirical systems/ML paper: reported accuracies are held-out measurements, not results forced by definition or self-citation.

full rationale

The paper’s load-bearing claim is empirical recognition accuracy of a dual-stream MobileViT fusion model on a held-out split of the synthetic JIVD dataset (Fig. 12–13, §VI.E), after training cGAN denoising and the classifier on the training split. That measurement is not algebraically equivalent to any fitted constant, uniqueness theorem, or renamed known law. Geometric PTZ control laws (26)–(30) are standard pinhole/Euler constructions from estimated pose; STFT mD extraction and MVDR/FFT parameter estimation are classical signal-processing steps, not self-definitional. Self-citation [6] only supplies target mesh/kinematics details used to build the simulator (“Specific details on the modeling process, kinematics, micro-motion parameters, and geometry of these targets can be found in [6]”); it does not define or force the reported 97.67% accuracy. Evaluation mismatches (oracle AirSim pose for PTZ vs. claimed ISAC estimates; LoS-only channels) are validity/correctness concerns, not circularity. No step reduces a claimed prediction to its inputs by construction.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 2 invented entities

Load-bearing content is an engineering pipeline plus synthetic evaluation. Physics/channel model and DL blocks are taken from standard literature; the paper’s own commitments are LoS-only simulation, co-located BS/PTZ, specific OFDM/array/STFT settings, and the dual-stream fusion design. No new physical entity is postulated. Free parameters are the usual trained weights and a few hand-set control/loss knobs (λ_loss, L_cls=40 px, SNR grid, architecture depths).

free parameters (5)
  • cGAN L1 weight λ_loss
    Balances adversarial vs pixel L1 loss in Eq. (46); value not numerically reported but controls denoising objective.
  • Recognition / denoise network weights (MobileViT-S/XS, U-Net G, PatchGAN D, YOLO) = trained on 80% sequences
    All end-to-end or stage-wise trained on JIVD train split; central accuracy depends on these fits.
  • L_cls = 40 pixels (required target width for FOV law) = 40
    Hand-set recognition pixel width entering adaptive FOV Eq. (30); directly sets zoom aggressiveness in dataset capture.
  • STFT window/step (L_win=128, A_step=4) and OFDM/frame geometry = L_win=128, A_step=4; f0=26 GHz, M=1024, N=4200, Ts=10 µs
    Hand-chosen time-frequency and waveform parameters that define the mD image the classifier sees.
  • mD feature projection dimension D_proj and fusion MLP
    Architectural size of ISAC branch projection before concat; not numerically specified but required for reported fusion.
axioms (6)
  • domain assumption Echo channel is a sum of K scattering points with standard far-field URA steering and complex fading (Eqs. 3–10).
    Standard MIMO radar/ISAC model invoked throughout §III; underpins position and mD extraction.
  • ad hoc to paper Only LoS paths are retained; NLoS is neglected for small targets at long range (§VI.A).
    Explicit simulation choice that removes urban multipath the introduction cites as a deployment concern.
  • domain assumption BS and PTZ camera are co-located at the origin with controllable pan/tilt/zoom matching estimated spherical angles and FOV law (Eqs. 26–30).
    Geometric cueing model in §II–III; required for “hundreds of meters” imaging claim.
  • ad hoc to paper AirSim visuals + Wireless InSite X3D LoS parameters substituted into the theoretical channel sufficiently represent real RF–vision pairs for accuracy claims.
    Entire empirical support is this synthetic stack (§VI); no real measurement calibration.
  • domain assumption STFT magnitude summed over range, energy-centered, yields discriminative micro-motion signatures across the ten subtypes (Fig. 3, §IV.A).
    Standard mD practice; paper treats distinct rotor vs flap patterns as given for classification.
  • standard math Cross-entropy on fused MobileViT features is an adequate training objective for fine-grained ID (§V.B).
    Routine supervised learning assumption.
invented entities (2)
  • JIVD (joint ISAC and vision dataset) no independent evidence
    purpose: Provide paired synchronized mD/echo and PTZ RGB samples across 10 subtypes, ranges, scenes, and weather for training/evaluation.
    New synthetic corpus generated by the authors; not an external benchmark. No independent public release or real-world twin is shown.
  • ISAC–vision fine-grained recognition network (dual MobileViT-S/XS + projection + MLP) no independent evidence
    purpose: Fuse denoised mD and YOLO-cropped RGB into 10-way subtype logits.
    Architectural composition of known backbones; “invented” only as this specific dual-stream wiring, with evidence solely from JIVD test accuracy.

pith-pipeline@v1.2.0-grok45-kimik3 · 24438 in / 4360 out tokens · 87173 ms · 2026-07-30T12:24:01.058181+00:00 · methodology

0 comments
read the original abstract

In this paper, we propose an integrated sensing and communications (ISAC) and vision fusion framework for fine-grained low-altitude target recognition. Specifically, we first utilize ISAC system to estimate the position of the low-altitude target. Then we adjust the working parameters of the Pan-TiltZoom (PTZ) camera based on the estimated target position, such that the camera can capture the image of tiny low-altitude target from several hundred meters away. After obtaining the wireless echo signal and visual image of low-altitude target, we employ the short-time Fourier transform (STFT) to obtain the micro-Doppler (mD) spectrum of the target from wireless echo signal, and design a conditional generative adversarial network (cGAN)-based denoising network to optimize the quality of the mD spectrum. Meanwhile, we employ YOLOv11 to detect the low-altitude target from visual image, and then crop the smallsized feature image of the target from the original image. Next, we design a fine-grained low-altitude target recognition network with MobileViT, which can fuse the optimized mD spectrum and the cropped feature image to distinguish the subcategory of low-altitude target. Moreover, we generate a joint ISAC and vision dataset (JIVD) for low-altitude target monitoring based on AirSim and Wireless InSite, which includes diverse target subcategories, scenarios, and weather conditions. The effectiveness and superiority of the proposed scheme have been demonstrated by simulation results.

Figures

Figures reproduced from arXiv: 2607.23789 by Boxuan Sun, Chuanbin Zhao, Feifei Gao, Hongliang Luo, Shengjie Quan, Zhonghua Chu.

Figure 1
Figure 1. Figure 1: Fine-grained low-altitude target recognition scenario based on ISAC [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The proposed ISAC and vision fusion framework for fine-grained low-altitude target recognition. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: mD spectrum of four different targets. (a) DJI MAVIC 3. (b) DJI [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: cGAN-based denoising network. B. Micro-Doppler Spectrum Denoising Enhancement In practical applications, the received echo signals are inevitably corrupted by noise. Consequently, under low signal￾to-noise ratio (SNR) conditions, the mD features become obscured and difficult to distinguish. Let us denote SNR as SNR = E h Hn,mx ∗ n,m [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Fine-grained low-altitude target recognition network with MobileViT. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Generation pipeline of joint ISAC and vision low-altitude target [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Composition of the genetated joint ISAC and vision low-alititude targets monitoring dataset. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of reconstruction quality metrics between noisy and denoised mD spectrum under different SNR. (a) MSE. (b) PSNR. (c) SSIM. 0 5 10 15 20 25 30 SNR (dB) 50 55 60 65 70 75 80 Acc (%) Denoised mD only Noisy mD only [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: F1-score performance under varying confidence threshold at IoU [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Histogram of IoU on test dataset 0 10 20 30 SNR (dB) 60 70 80 90 100 Acc (%) Denoised mD only Noisy mD and vision Denoised mD and vision Vision only [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Testing recognition accuracy versus SNRs of different schemes [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: UMAP analysis of 30 dB test data significantly at low SNRs with an average accuracy of 86.25%, which requires a high SNR of 30 dB to eventually match the performance of the proposed method. Meanwhile, the single￾modal schemes exhibit significantly inferior performance. The average accuracy of the “Vision only” scheme is limited to 74.34%, while the “Denoised mD only” scheme achieves only 71.08%. The subst… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 5 linked inside Pith

  1. [1]

    6G non-terrestrial networks enabled low-altitude economy: Opportunities and challenges,

    Y . Jianget al., “6G non-terrestrial networks enabled low-altitude economy: Opportunities and challenges,”arXiv e-prints, p. arXiv:2311.09047, Nov. 2023. 13

    Pith/arXiv arXiv 2023

  2. [2]

    The potential of low-altitude airspace: The future of urban air transportation,

    H. Huang, J. Su, and F.-Y . Wang, “The potential of low-altitude airspace: The future of urban air transportation,”IEEE Trans. Intell. V eh., vol. 9, no. 8, pp. 5250–5254, Aug. 2024

    2024

  3. [3]

    Non-terrestrial networks in the 6G era: Challenges and opportunities,

    M. Giordani and M. Zorzi, “Non-terrestrial networks in the 6G era: Challenges and opportunities,”IEEE Network, vol. 35, no. 2, pp. 244– 251, Mar. 2021

    2021

  4. [4]

    IoTaaS: drone-based internet of things as a service framework for smart cities,

    M. A. Hoqueet al., “IoTaaS: drone-based internet of things as a service framework for smart cities,”IEEE Internet Things J., vol. 9, no. 14, pp. 12 425–12 439, Jul. 2022

    2022

  5. [5]

    Networked ISAC- based UA V tracking and handover toward low-altitude economy,

    C. Zhao, Y . Feng, H. Luo, F. Gao, F. Liu, and S. Jin, “Networked ISAC- based UA V tracking and handover toward low-altitude economy,”IEEE Trans. Wireless Commun., vol. 24, no. 9, pp. 7670–7685, Apr. 2025

    2025

  6. [6]

    AirGuard: UA V and bird recognition scheme for integrated sensing and communications system,

    H. Luoet al., “AirGuard: UA V and bird recognition scheme for integrated sensing and communications system,”IEEE J. Sel. Areas Commun., vol. 44, pp. 835–848, 2026

    2026

  7. [7]

    Toward 6G with connected sky: UA Vs and beyond,

    M. Mozaffari, X. Lin, and S. Hayes, “Toward 6G with connected sky: UA Vs and beyond,”IEEE Commun. Mag., vol. 59, no. 12, pp. 74–80, Jan. 2022

    2022

  8. [8]

    Enhanced UA V detection and classification using machine learning and MIMO radars,

    A. N. Sayedet al., “Enhanced UA V detection and classification using machine learning and MIMO radars,”IEEE Trans. on Microw. Theory and Tech., vol. 72, no. 11, pp. 6716–6727, Nov. 2024

    2024

  9. [9]

    A method of photoelectric capture of UA V under rough guidance,

    B. Maet al., “A method of photoelectric capture of UA V under rough guidance,” inProc. Int. Conf. Cloud Comput. Big Data Anal., Chengdu, China, Apr. 2024, pp. 184–187

    2024

  10. [10]

    Low-altitude protection technology of anti-UA Vs based on multisource detection information fusion,

    S. Chenet al., “Low-altitude protection technology of anti-UA Vs based on multisource detection information fusion,”Int. J. Adv. Robot. Syst., vol. 17, no. 5, p. 1729881420962907, 2020

    2020

  11. [11]

    Amateur drone monitoring: State- of-the-art architectures, key enabling technologies, and future research directions,

    Z. Kaleem and M. H. Rehmani, “Amateur drone monitoring: State- of-the-art architectures, key enabling technologies, and future research directions,”IEEE Wireless Commun., vol. 25, no. 2, pp. 150–159, May 2018

    2018

  12. [12]

    Domain-adaptive UA V recognition using IR-UWB radar and a lightweight Mamba-based network,

    S. Liet al., “Domain-adaptive UA V recognition using IR-UWB radar and a lightweight Mamba-based network,”IEEE Sensors J., vol. 25, no. 18, pp. 34 913–34 926, Aug. 2025

    2025

  13. [13]

    Micro-Doppler mini-UA V classification using empirical-mode decomposition features,

    B.-S. Ohet al., “Micro-Doppler mini-UA V classification using empirical-mode decomposition features,”IEEE Geosci. and Remote Sens. Lett., vol. 15, no. 2, pp. 227–231, Dec. 2018

    2018

  14. [14]

    Classification of birds and UA Vs based on radar polarimetry,

    B. Torvik, K. E. Olsen, and H. Griffiths, “Classification of birds and UA Vs based on radar polarimetry,”IEEE Geosci. and Remote Sens. Lett., vol. 13, no. 9, pp. 1305–1309, Jul. 2016

    2016

  15. [15]

    What is YOLOv5: A deep look into the internal features of the popular object detector,

    R. Khanam and M. Hussain, “What is YOLOv5: A deep look into the internal features of the popular object detector,”arXiv e-prints, p. arXiv:2407.20892, Jul. 2024

    Pith/arXiv arXiv 2024

  16. [16]

    Drone detection and tracking: An edge-deployable efficient algorithm based on vision sensor,

    B. Wang, J. Li, M. Zhou, and Q. Lu, “Drone detection and tracking: An edge-deployable efficient algorithm based on vision sensor,”IEEE Sensors J., vol. 25, no. 17, pp. 34 126–34 140, Jul. 2025

    2025

  17. [17]

    A review on YOLOv8 and its advancements,

    M. Sohan, T. Sai Ram, and C. V . Rami Reddy, “A review on YOLOv8 and its advancements,” inProc. Int. Conf. Data Intell. Cogn. Inform., 2024, pp. 529–545

    2024

  18. [18]

    Experiments on vision-based methods to detect, identify and track UA Vs using thermal camera,

    D. Dosi, C. E. Abraham, and D. Kumar, “Experiments on vision-based methods to detect, identify and track UA Vs using thermal camera,” in Proc. Int. Conf. Emerg. Technol. Auton. Aer . V eh., Bangalore, India, Aug. 2025, pp. 1–6

    2025

  19. [19]

    Drone detection using convolutional neural networks,

    F. Mahdavi and R. Rajabi, “Drone detection using convolutional neural networks,” inProc. Iranian Conf. Signal Process. Intell. Syst., Mashhad, Iran, Dec. 2020, pp. 1–5

    2020

  20. [20]

    Machine learning-based drone detection and classification: State-of-the-art in research,

    B. Taha and A. Shoufan, “Machine learning-based drone detection and classification: State-of-the-art in research,”IEEE Access, vol. 7, pp. 138 669–138 682, Sep. 2019

    2019

  21. [21]

    Real-time UA V and payload detection and classification system using radar and camera sensor fusion,

    V . Mehtaet al., “Real-time UA V and payload detection and classification system using radar and camera sensor fusion,” inProc. IEEE/AIAA Digit. Avion. Syst. Conf., Barcelona, Spain, 2023, pp. 1–6

    2023

  22. [22]

    Edge AI-enabled radar and camera integration for real-time drone detection and classification,

    V . Mehtaet al., “Edge AI-enabled radar and camera integration for real-time drone detection and classification,” inProc. IEEE/AIAA Digit. Avion. Syst. Conf., Montreal, Canada, 2025, pp. 1–7

    2025

  23. [23]

    Radar-optical fusion detection of UA V based on improved YOLOv7-tiny,

    H. Tanget al., “Radar-optical fusion detection of UA V based on improved YOLOv7-tiny,”Meas. Sci. Technol., vol. 35, no. 8, p. 085110, May 2024

    2024

  24. [24]

    Perceptive mobile networks: Cellular networks with radio vision via joint communication and radar sensing,

    A. Zhanget al., “Perceptive mobile networks: Cellular networks with radio vision via joint communication and radar sensing,”IEEE V eh. Technol. Mag., vol. 16, no. 2, pp. 20–30, Jun. 2021

    2021

  25. [25]

    Integrated sensing and communications: Toward dual-functional wire- less networks for 6G and beyond,

    F. Liu, Y . Cui, C. Masouros, J. Xu, T. X. Han, Y . C. Eldar, and S. Buzzi, “Integrated sensing and communications: Toward dual-functional wire- less networks for 6G and beyond,”IEEE J. Sel. Areas Commun., vol. 40, no. 6, pp. 1728–1767, Jun. 2022

    2022

  26. [26]

    Toward 6G networks: Use cases and technologies,

    M. Giordani, M. Polese, M. Mezzavilla, S. Rangan, and M. Zorzi, “Toward 6G networks: Use cases and technologies,”IEEE Commun. Mag., vol. 58, no. 3, pp. 55–61, Mar. 2020

    2020

  27. [27]

    Integrated sensing and communications framework for 6G networks,

    H. Luo, T. Zhang, C. Zhao, Y . Wang, B. Lin, Y . Jiang, D. Luo, and F. Gao, “Integrated sensing and communications framework for 6G networks,”IEEE Wireless Commun., vol. 32, no. 6, pp. 102–109, Dec. 2025

    2025

  28. [28]

    Convergent communication, sensing and localiza- tion in 6G systems: An overview of technologies, opportunities and challenges,

    C. De Limaet al., “Convergent communication, sensing and localiza- tion in 6G systems: An overview of technologies, opportunities and challenges,”IEEE Access, vol. 9, pp. 26 902–26 925, Jan. 2021

    2021

  29. [29]

    Seven defining features of terahertz (THz) wireless sys- tems: A fellowship of communication and sensing,

    C. Chaccour, M. N. Soorki, W. Saad, M. Bennis, P. Popovski, and M. Debbah, “Seven defining features of terahertz (THz) wireless sys- tems: A fellowship of communication and sensing,”IEEE Commun. Surv. Tut., vol. 24, no. 2, pp. 967–993, Jan. 2022

    2022

  30. [30]

    6G wire- less communication systems: Applications, requirements, technologies, challenges, and research directions,

    M. Z. Chowdhury, M. Shahjalal, S. Ahmed, and Y . M. Jang, “6G wire- less communication systems: Applications, requirements, technologies, challenges, and research directions,”IEEE Open J. Commun. Soc., vol. 1, pp. 957–975, Jul. 2020

    2020

  31. [31]

    Cooperative ISAC-empowered low-altitude economy,

    J. Tang, Y . Yu, C. Pan, H. Ren, D. Wang, J. Wang, and X. You, “Cooperative ISAC-empowered low-altitude economy,”IEEE Trans. Wireless Commun., vol. 24, no. 5, pp. 3837–3853, May 2025

    2025

  32. [32]

    Learned off-grid imager for low-altitude economy with cooperative ISAC network,

    Y . Huang, J. Yang, S. Xia, C.-K. Wen, and S. Jin, “Learned off-grid imager for low-altitude economy with cooperative ISAC network,”IEEE Trans. Wireless Commun., vol. 25, pp. 3333–3348, 2026

    2026

  33. [33]

    UA V’s rotor micro-doppler feature extraction using integrated sensing and communication signal: Algorithm design and testbed evaluation,

    J. Wei, D. Ma, F. He, Q. Zhang, Z. Feng, Z. Liu, and T. Liang, “UA V’s rotor micro-doppler feature extraction using integrated sensing and communication signal: Algorithm design and testbed evaluation,” IEEE Trans. Wireless Commun., vol. 24, no. 12, pp. 10 166–10 182, Jul. 2025

    2025

  34. [34]

    DC-Former network empowered UA V and bird recognition based on integrated sensing and communi- cation system,

    J. Xue, Q. Zhang, D. Ma, and J. Wei, “DC-Former network empowered UA V and bird recognition based on integrated sensing and communi- cation system,” inProc. Int. Conf. Comput. Commun. Syst., Chengdu, China, Apr. 2025, pp.927-932

    2025

  35. [35]

    Ml joint CFO and channel estimation in OFDM systems with timing ambiguity,

    J. Chen, Y .-c. Wu, S. Ma, and T.-s. Ng, “Ml joint CFO and channel estimation in OFDM systems with timing ambiguity,”IEEE Trans. Wireless Commun., vol. 7, no. 7, pp. 2436–2440, Jul. 2008

    2008

  36. [36]

    Generative adversarial nets,

    I. J. Goodfellowet al., “Generative adversarial nets,” inAdv. Neural Inf. Process. Syst., vol. 27, 2014

    2014

  37. [37]

    Yolov11: An overview of the key architectural enhancements,

    R. Khanam and M. Hussain, “Yolov11: An overview of the key architectural enhancements,”arXiv e-prints, p. arXiv:2410.17725, Oct. 2024

    Pith/arXiv arXiv 2024

  38. [38]

    MobileViT: Light-weight, general- purpose, and mobile-friendly vision transformer,

    S. Mehta and M. Rastegari, “MobileViT: Light-weight, general- purpose, and mobile-friendly vision transformer,”arXiv e-prints, p. arXiv:2110.02178, Oct. 2021

    Pith/arXiv arXiv 2021

  39. [39]

    DoA estimation and capacity analysis for 3-D millimeter wave massive-MIMO/FD-MIMO OFDM systems,

    R. Shafin, L. Liu, J. Zhang, and Y .-C. Wu, “DoA estimation and capacity analysis for 3-D millimeter wave massive-MIMO/FD-MIMO OFDM systems,”IEEE Trans. Wireless Commun., vol. 15, no. 10, pp. 6963– 6978, Oct. 2016

    2016

  40. [40]

    Integrated sensing and communications for V2I networks: Dynamic predictive beamforming for extended vehicle targets,

    Z. Du, F. Liu, W. Yuan, C. Masouros, Z. Zhang, S. Xia, and G. Caire, “Integrated sensing and communications for V2I networks: Dynamic predictive beamforming for extended vehicle targets,”IEEE Trans. Wireless Commun., vol. 22, no. 6, pp. 3612–3627, Jun. 2023

    2023

  41. [41]

    Broad beam designs for broadcast channels,

    C. Du and Y . Jiang, “Broad beam designs for broadcast channels,”IEEE Trans. Signal Process., vol. 72, pp. 3819–3833, Aug. 2024

    2024

  42. [42]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” inProc. Int. Conf. Med. Image Comput. Comput.-Assist. Interv., 2015, pp. 234–241

    2015

  43. [43]

    Image-to-image translation with conditional adversarial networks,

    P. Isola, J.-Y . Zhu, T. Zhou, and A. A. Efros, “Image-to-image translation with conditional adversarial networks,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., 2017, pp. 5967–5976

    2017

  44. [44]

    AirSim: High-fidelity visual and physical simulation for autonomous vehicles,

    S. Shah, D. Dey, C. Lovett, and A. Kapoor, “AirSim: High-fidelity visual and physical simulation for autonomous vehicles,” inProc. Int. Conf. Field Serv. Robot., 2017, pp. 621–635

    2017

  45. [45]

    B. O. Community,Blender - a 3D modelling and rendering package, Blender Foundation, Stichting Blender Foundation, Amsterdam, 2018. [Online]. Available: http://www.blender.org

    2018

  46. [46]

    Wireless insite

    Remcom, “Wireless insite.” [Online]. Available: https://www.remcom. com/wireless-insite-em-propagation-software

  47. [47]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE Trans. Image Process., vol. 13, no. 4, pp. 600–612, Apr. 2004

    2004

  48. [48]

    Umap: Uniform manifold approximation and projection for dimension reduction,

    L. McInnes, J. Healy, and J. Melville, “Umap: Uniform manifold approximation and projection for dimension reduction,”arXiv e-prints, p. arXiv:1802.03426, Sep. 2018

    Pith/arXiv arXiv 2018

  49. [49]

    Support vector machines,

    M. Hearst, S. Dumais, E. Osuna, J. Platt, and B. Scholkopf, “Support vector machines,”IEEE Intell. Syst. Their Appl., vol. 13, no. 4, pp. 18– 28, Aug. 1998

    1998