Work

TinyML-enabled LoRaWAN acoustic monitoring for illegal logging detection

Year: 2026

Type: article

Abstract: Illegal logging remains a leading driver of forest degradation, particularly in remote regions where limited connectivity and strict energy budgets constrain conventional monitoring approaches that depend on continuous network access, high-power hardware, or centralized processing. Existing acoustic and Internet of Things (IoT)-based detection systems are typically validated in a single environment and rarely combine acoustic detection with environmental context sensing, which limits their demonstrated generalizability and situational awareness. To address these gaps, this paper presents a Tiny Machine Learning (TinyML)-enabled Long Range Wide Area Network (LoRaWAN) forest monitoring system that performs real-time, on-device acoustic detection of chainsaw activity while acquiring temperature, humidity, soil-moisture, and flame readings as supplementary environmental context. Acoustic inference is executed on the sensing node by a lightweight MobileNetV2-based model with post-training int8 quantization, and a sliding-window segmentation strategy combined with rolling decision logic suppresses transient false detections to support stable event-driven operation. Int8 quantization reduced inference latency to 74 ms and peak Random Access Memory (RAM) usage to 177.1 KB while preserving a classification accuracy of 94.78%. The complete system was validated in two climatically distinct 7-day field deployments, in a tropical forest and in a semi-arid environment, which confirmed reliable end-to-end operation and yielded reliable communication ranges of 400 m and 1200 m, respectively, quantifying the influence of vegetation density on signal propagation. These results demonstrate the feasibility of combining on-device TinyML inference, environmental context sensing, and event-driven long-range communication for real-time illegal logging detection across climatically diverse environments.

Source: Journal of Engineering Research

Authors: Abdulsalam A. Almathhachi, Nor Fadzilah Abdullah, Haider A.H. Alobaidy, Asma Abu-Samah

Institutions: National University of Malaysia, Nahrain University

Language: English

Topic: IoT Networks and Protocols

Subfield: Electrical and Electronic Engineering

Field: Engineering

Domain: Physical Sciences

SDGs: Life on land, Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $1,100

Funder: Ministry of Higher Education, Malaysia