How Anti-UAV Radars Accurately Distinguish Drones from Birds

Introduction

In low-altitude security scenarios such as ports, oil & gas depots, airports and industrial parks, mini UAVs and birds both fall under the category of "low, slow, small targets". They feature small sizes, low flight altitudes and slow moving speeds. Conventional radars are prone to misclassify birds as intruding UAVs, generating thousands of false alarms every day. This significantly increases the 24/7 security workload and may even obscure real threats posed by unauthorized drones.

Our 2D active phased array low-altitude surveillance radars and LW-Ku205 airborne Ku-band radars adopt micro-Doppler feature recognition as the core technology, supplemented by multi-dimensional judgment criteria including RCS scattering, track kinematics, AI deep learning and optoelectronic fusion. These solutions can reduce bird-caused false alarm rates by over 90%, enabling all-weather high-precision target classification, and serve as core sensing equipment for low-altitude security of critical infrastructure.

I. Core Identification Criterion: Micro-Doppler Effect (The Most Critical Differentiation Method)

Besides capturing the overall flight speed of targets, radars can detect subtle frequency fluctuations induced by tiny moving components on targets. This echo frequency modulation phenomenon is defined as the micro-Doppler effect. Rotating UAV propellers and slowly flapping bird wings follow completely different motion mechanisms, leading to drastically different spectral signatures. This forms the fundamental basis for radars to distinguish the two types of targets.

1. Unique Spectral Features of Multi-Rotor UAVs

1. High-frequency, Symmetrical and Regular Harmonic Stripes Driven by motors, UAV propellers rotate hundreds of times per second with a rotational frequency of 50~100Hz. Synchronized rotation of multi-rotors on quadcopters and hexacopters generates dense, evenly symmetrical light-dark blade flash stripes on spectrograms with strong periodicity and high identifiability.
2. Exclusive Hover Signature UAVs can hover stably in the air for a long time. While the overall translational speed of the fuselage approaches zero, propellers keep spinning at high speed, producing distinct micro-Doppler spectra captured by radars. Birds cannot maintain stable hovering, and this feature can quickly filter out most bird targets.
3. Wide Spectral Bandwidth The maximum micro-Doppler frequency of UAVs can reach several thousand hertz, covering a far wider spectral range than birds with clearly defined feature boundaries on time-frequency diagrams.

2. Spectral Features of Flapping Bird Wings

1. Low-frequency, Irregular and Asymmetric Fluctuations Bird wings flap at a frequency of only 4~10Hz, producing gentle, irregular wavy curves without dense harmonic lines on spectra. When gliding with stationary wings, the micro-Doppler feature almost disappears entirely.
2. Constantly Shifting Spectra During Flight The wing flapping rhythm of birds keeps changing during gliding, circling and diving, resulting in continuously deformed spectral patterns without fixed stable waveforms.
3. Narrow Spectral Bandwidth The maximum Doppler frequency of birds stays below 200Hz, forming an obvious spectral gap from UAVs. Algorithms can quickly set thresholds to separate the two categories.

II. Four Auxiliary Verification Dimensions to Eliminate Misjudgment via Multi-Layer Logic

Reliance on micro-Doppler features alone may lead to rare misclassification. Industrial-grade security radars integrate four layers of feature joint judgment with cross-verification to guarantee target recognition accuracy.

(1) RCS (Radar Cross Section) Material Difference

• UAVs: The fuselage, motors and propellers are made of carbon fiber and metal components, generating multiple discrete strong reflection peaks on echoes. Mini racing drones maintain an RCS range of 0.01~0.1 ㎡, with fixed reflection highlights formed by metal parts.
• Birds: Composed of flesh and feathers with nearly no metal reflective structures, birds deliver soft and single integrated echo signals without local strong reflection points. Their RCS values fluctuate within a wider and weaker overall range.

(2) Kinematic Logic Judgment of Flight Tracks

Radars continuously track targets for 3 to 10 seconds, recording speed, altitude, turning angle and dwell duration to distinguish target types based on motion patterns:

1. Motion Characteristics of UAVs UAVs can fly straight at a constant speed over long distances, hover in fixed positions and make precise sharp turns. They can maintain a steady altitude for extended periods and move along regular linear paths such as waterways, tank walls and perimeter fences with stable and controllable flight routes.
2. Motion Characteristics of Birds Birds fly without fixed routes, circling randomly with frequent altitude shifts and speed changes. They cannot stay stationary for long periods, swerve arbitrarily when encountering obstacles, and mostly travel in scattered flocks without cruising along fixed paths for long distances.

(3) Target Size and Altitude Range Screening

• UAVs have fixed dimensions ranging from 20cm to 2m, capable of sustained flight at any altitude between 5m and 1000m.
• Birds vary drastically in body size. Seabirds and migratory birds mostly fly close to the sea or ground, rarely cruising steadily at altitudes of hundreds of meters.

(4) Secondary Cross-Verification via Radar & Optoelectronic Fusion

After radars output target coordinates, the system automatically links visible-light / infrared optoelectronic pan-tilts to lock onto targets for secondary confirmation through thermal imaging signatures:

1. UAVs: Motors and electronic speed controllers continuously generate heat, showing multiple dotted high-temperature heat sources with angular and distinct fuselage contours on thermal images.
2. Birds: As constant-temperature living creatures, birds feature evenly distributed body heat forming a smooth, unified single heat source without local high-temperature spots.

III. Complete AI Intelligent Recognition Workflow of Radars

1. Echo Clutter Preprocessing The pulse Doppler architecture filters static clutter generated by sea waves, metal tank bodies and buildings, extracting raw echo signals of moving aerial targets.
2. Time-Frequency Spectrogram Generation Short-Time Fourier Transform is applied to parse echoes and generate exclusive micro-Doppler spectrograms.
3. Automatic Extraction of Multi-Dimensional Features Algorithms extract five core parameters: spectral bandwidth, harmonic quantity, symmetry, RCS intensity and flight track kinematics.
4. Automatic Classification via Deep Learning Model A CNN neural network trained with hundreds of thousands of measured UAV and bird samples automatically identifies target categories and outputs recognition confidence levels.
5. Secondary Correction via Long-Time-Series Flight Tracks Continuous track data recorded over dozens of seconds correct misjudgments caused by accidental short-term spectral deviations, filtering out temporary abnormal feature readings.
6. Hierarchical Alarm Output Targets identified as UAVs trigger high-risk alarms with complete flight track logs archived; bird targets are filtered directly without generating invalid alarm records.