Lumber Guardian – Edge AI Solution

Industry: Environmental Protection

Technologies and Processes: Custom Hardware Development, Edge AI, LoRaWAN Connectivity, Solar-powered IoT Devices

Illegal logging is damaging forests across the world, with Croatia being one of the most affected countries in the European Union. To address this challenge, our Innovation department developed the Lumber Guardian, an edge AI-based system designed to detect chainsaw sounds and alert authorities via LoRaWAN connectivity. This device minimizes the need for human monitoring, offering an eco-friendly and cost-effective solution for forest conservation.

Lumber Guardian is a “set-and-forget” smart device that detects illegal logging activities through real-time sound classification once it is installed on a tree. Equipped with AI capabilities, it processes audio locally using a pattern-based neural network to identify chainsaw sounds with over 95% accuracy. When a chainsaw sound is confirmed, the device sends a report to a server, enabling authorities to respond effectively.

Beyond chainsaw detection, the system can be repurposed for various sound-monitoring applications, making it a versatile tool for diverse use cases.

Our team designed and produced custom hardware to support the Lumber Guardian’s requirements. The device’s hardware is centered around the STMicroelectronics STM32H7A3VI microcontroller, supported by components such as:

• Audio input: Five microphones for sound localization and activity detection
• Connectivity: LoRaWAN modem for long-range communication
• Power management: Solar-powered battery with a maximum power point tracking charger for efficiency

Key Features:

• ZephyrOS-based firmware for streamlined development
• Real-time sound classification using double buffering
• Solar-powered, low-energy operation for long-term deployment
• Easy installation with no setup required
• Additional functionalities, including temperature and humidity readings

The system also employs advanced digital signal processing techniques (MFE and MFCC) to extract the most relevant features from sound data. This is further enhanced by a Bayesian probability-based scoring system to reduce false positives and conserve energy.