Ultra-Early Off-Grid Wildfire & Ember Acoustic-Visual Detection Node

• Contest: The Resilient America Preparedness Challenge
• Track: Track A — Developers
• Primary Focus Areas: Anticipate and Mitigate Risk / Operate in Mission-Critical Environments
• Core Tech Stack: Arduino® UNO™ Q (Qualcomm® Dragonwing™ Platform + STM32U585 MCU), Edge Impulse, FLIR Lepton 3.5, LoRa SX1262 Mesh

Wildfires present a rapidly escalating crisis to infrastructure and human life, yet existing early warning systems remain fundamentally flawed. Satellite thermal passes can take hours to update, while traditional visual cameras are blinded by night conditions, dense forest canopies, and dust clouds. Crucially, when wildland fires break out, power grids and cellular infrastructure are often destroyed first, creating catastrophic communication blackouts.

EmberSentry-Edge is a self-contained, solar-powered multi-modal edge intelligence node engineered for deployment along high-risk wilderness-urban interfaces (WUI). Powered by the Arduino® UNO™ Q dual-processor platform and optimized via Edge Impulse, EmberSentry-Edge detects wildfires in the pre-ignition/pyrolysis stage (before visible flame or massive smoke plumes emerge) using an array of environmental, acoustic, and radiometric thermal sensors. Operating with zero dependence on cloud connectivity, cellular networks, or grid power, EmberSentry-Edge forms an autonomous, self-healing off-grid LoRa mesh to route instant threat vectors directly to first responders.

The core architectural innovation of EmberSentry-Edge relies on the Arduino UNO Q’s dual-brain capability to execute a multi-layered, duty-cycled AI pipeline.

• Execution Environment: Bare-metal / Zephyr OS on the low-power STM32 MCU.
• Hardware Core: Driven by the low-power MCU domain of the Arduino UNO Q platform.
• Sensor Suite: Sensirion SEN55 / Bosch BME688 (VOCs, CO, Particulate Matter PM2.5/10) + BME280 (Temperature, Humidity, Barometric Pressure).
• Edge Impulse Model: 1D-Convolutional Neural Network (Conv1D) running continuous time-series anomaly detection.
• Function: Continuously monitors the local atmosphere for chemical pyrolysis signatures (gas spikes emitted as wood heat decays prior to combustion). When an anomaly score breaches set thresholds, it asserts a hardware GPIO line to wake the Qualcomm MPU.

• Execution Environment: Debian Linux on the high-performance Qualcomm Dragonwing QRB2210.
• Sensor Suite: FLIR Lepton 3.5 Radiometric LWIR thermal camera (160×120), OV5640 5MP optical camera (MIPI-CSI), and Knowles SPH0645LM4H digital MEMS microphone.
• Edge Impulse Models:

1. Thermal/Visual Model: Edge Impulse FOMO (Faster Objects, More Objects) trained on radiometric heat gradients and smoke patterns to locate heat centroids and hot spots (>150°C) through dense canopy or total darkness.
2. Acoustic Model: MFE (Mel-Filterbank Energy) audio classifier trained to recognize the acoustic frequencies of dry timber cracking and canopy fire noise.

• Function: Validates the event with high confidence, preventing false alarms, and constructs a signed emergency payload containing geo-location, temperature delta, and wind vectors.

The system utilizes a high-efficiency hybrid UART + GPIO Hardware Interrupt scheme to connect the two onboard processing domains:

• Wake Signaling: The STM32 MCU drives a dedicated physical GPIO line HIGH to trigger the Qualcomm PMIC when an atmospheric pyrolysis event is detected.
• Data Transport: Telemetry payloads (sensor metrics and trigger states) transfer across the internal UART bus running at 921,600 baud using the Arduino Bridge / RPC Protocol wrapped in MessagePack.

EmberSentry-Edge provides two-tier off-grid connectivity via an onboard Semtech SX1262 LoRa transceiver:

1. Local Node-to-Node Mesh (RadioHead on STM32): Low-overhead, raw packet transmission carrying binary sensor payloads between neighbor nodes.
2. Long-Haul Gateway Relay (Reticulum Network Stack on Qualcomm MPU): Cryptographically signed, dynamic mesh routing running via Python on Linux to route critical alerts and thermal metadata across long distances to the Incident Command Post (ICP).

To guarantee operational autonomy through cloud cover and heavy smoke haze, the system utilizes a 12.8V 6.0Ah (76.8 Wh) LiFePO4 battery paired with a 10W monocrystalline solar panel managed by a CN3791 MPPT Solar Charger.

Daily Energy Consumption:

• Baseline Sensing: 24 hours × 0.014 Watts = 0.336 Wh / day
• Layer 2 Verification (10 events/day including FLIR Lepton active draw + shutter FFC burst): 10 × 60 seconds × 2.65 Watts = 0.442 Wh / day
• LoRa Alerts (10 transmissions/day): 10 × 5 seconds × 4.25 Watts = 0.059 Wh / day
• Total Daily Energy Requirement: ~1.12 Wh / day (accounting for power conversion efficiency)

Autonomy Reserve (Zero Solar Input):

• Battery Capacity: 76.8 Wh (12.8V @ 6.0Ah)
• Maximum No-Sun Operational Window: 76.8 Wh ÷ 1.12 Wh/day = 68.5 Days

EmberSentry-Edge remains operational across all phases of a wildfire event:

• Mitigation (Pre-Ignition): Detects early pyrolysis gas emissions and timber stress prior to flame production.
• Containment (Active Fire): FLIR Lepton thermal imaging penetrates smoke plumes to spot embers crossing firebreaks and calculate real-time Rate-of-Spread (RoS) vectors.
• Recovery (Mop-Up Phase): Identifies subterranean root fires and sub-surface hot spots (>150°C) beneath cool top-ash to prevent reignition.

The Bill of Materials balances high-performance multi-modal sensing with commercial viability for field deployment at scale. Component documentation and technical specification links are included below.

• Innovation (35%): Multi-modal fusion of gas, acoustics, optical, and radiometric thermal data on a duty-cycled dual-brain architecture.
• Prototype Functionality (30%): Complete integration of Arduino UNO Q hardware, Edge Impulse ML models, and off-grid LoRa mesh protocols.
• Resilience Impact (25%): Ultra-early detection combined with complete off-grid autonomy during infrastructure destruction.
• Team Potential (10%): Leverages established open-source hardware standards (Bosch, Sensirion, FLIR) and software stacks (Zephyr OS, Reticulum, Edge Impulse).