Send Less, Save More: Energy-Efficiency Benchmark of Embedded CNN Inference vs. Data Transmission in IoT

TL;DR

This work benchmarks energy efficiency for IoT environmental monitoring by comparing cloud-based image processing against on-device CNN inference on ESP32-S3, paired with LPWAN transmissions. It evaluates two lightweight CNNs (MobileNetV2, SqueezeNet) trained and quantized for edge deployment on two domain datasets (PlantVillage and Cub), across multiple network protocols (LTE-M, NB-IoT, LoRaWAN) and sending strategies (image vs. inference results). Key findings show that edge inference with result-based transmission can reduce node energy by up to a factor of $5$, and that strategic scheduling (sending results only every few cycles) can yield up to about $14\times$ additional energy savings, significantly extending device lifetimes in remote settings. The results offer practical guidance on model choice, quantization, and network configurations to minimize power use in autonomous environmental monitoring deployments.

Abstract

The integration of the Internet of Things (IoT) and Artificial Intelligence offers significant opportunities to enhance our ability to monitor and address ecological changes. As environmental challenges become increasingly pressing, the need for effective remote monitoring solutions is more critical than ever. A major challenge in designing IoT applications for environmental monitoring - particularly those involving image data - is to create energy-efficient IoT devices capable of long-term operation in remote areas with limited power availability. Advancements in the field of Tiny Machine Learning allow the use of Convolutional Neural Networks (CNNs) on resource-constrained, battery-operated microcontrollers. Since data transfer is energy-intensive, performing inference directly on microcontrollers to reduce the message size can extend the operational lifespan of IoT nodes. This work evaluates the use of common Low Power Wide Area Networks and compressed CNNs trained on domain specific datasets on an ESP32-S3. Our experiments demonstrate, among other things, that executing CNN inference on-device and transmitting only the results reduces the overall energy consumption by a factor of up to five compared to sending raw image data. These findings advocate the development of IoT applications with reduced carbon footprint and capable of operating autonomously in environmental monitoring scenarios by incorporating EmbeddedML.

Figures (9)

• Figure 1: Monitoring crop health of cultivated plants
• Figure 2: Identifying/counting birds for biodiversity monitoring
• Figure 4: Energy consumption of an iot node for transmitting raw image data versus on-device CNN classification inference followed by transmitting the classified result by different network protocols.
• Figure 5: Scenario Design for Cloud-ML and Embedded-ML. Devices capture images intermittently, then either transmit as JPEG via cellular network (Cloud-ML) or perform local CNN inference and transmit the result if relevant (Embedded-ML), before returning to sleep mode.
• Figure 6: Overview of the model preparation pipeline for embedded ml deployment.