Technology - Lira Intelligence | 8 Core AI Technologies

8 Core Technologies

Self-developed, industry-leading, third-party verified

👁️

AI Vision Recognition

98.5% Lab / 96.2% On-site

YOLOv8 custom model, 50+ categories, <30ms latency, third-party verified.

YOLOv8m/l custom architecture
C2f module optimization
Focal Loss function
50+ categories training data

🎯

Small Object Detection

95% Lab / 88% On-site (3mm)

4-scale FPN, P2 small object layer, CBAM attention, 2 patents.

P2 detection layer (small objects)
4-scale FPN structure
CBAM attention mechanism
Small object weight enhancement

🌈

Multi-Spectral Recognition

98% Lab / 95% On-site

NIR/SWIR spectral analysis, visible light fusion, PET/PVC/PE sorting, 2 patents.

NIR (900-1700nm)
SWIR (1700-2500nm)
Visible light fusion
Material classification algorithm

📐

3D Vision

Depth Error <2mm

Structured light/ToF depth camera, PointNet++ processing, irregular object volume measurement.

Structured light/ToF depth
PointNet++ segmentation
3D detection (VoteNet)
Instance segmentation (3D-MPA)

🤖

Robotics Control

60/min Lab / 52/min On-site

ROS 2 + MoveIt, RRT* planning, visual servo ±0.1mm, 6-robot collaboration.

ROS 2 framework
RRT* motion planning
Visual servo (Eye-in-Hand)
6-axis force control impedance

⚡

Edge Computing

62 FPS (Jetson Orin NX INT8)

TensorRT INT8 acceleration, zero-copy pipeline, multi-CUDA stream, 1/5 power consumption, 5 copyrights.

TensorRT FP16/INT8
Zero-copy GPU pipeline
Multi-CUDA stream concurrency
Power optimization (1/5 x86)

🔗

IoT Device Management

10K+ Device Concurrency

MQTT + Redis message bus, remote monitoring, OTA updates, predictive maintenance.

MQTT protocol
Redis message queue
Device status monitoring
OTA remote updates

📊

Predictive Analytics

MAPE <10%

LSTM+Prophet hybrid model, recycling volume prediction, price trend analysis, maintenance alerts.

LSTM time-series prediction
Prophet trend analysis
Multi-factor fusion
Automatic model selection

Core Technology