Energy Efficient LoRa-Based AIoT Setup for Human Movement Classification

Abstract

In healthcare, artificial intelligence of things (AIoT) enhances patient care, diagnostics, and operational efficiency by integrating intelligent data analysis with medical devices. This research develops an energy-efficient AIoT setup for real-time hand movement categorization. The proposed setup consists of an edge device (Arduino Nano 33 BLE Sense Rev2 and Long Range [LoRa] Ra-02), a gateway (ESP-32 and LoRa Ra-02), and the ThingSpeak Internet of Things (IoT) platform. The setup is made energy efficient by characterizing its performance with and without obstacles between the edge device and the gateway. For each such scenario, the LoRa parameters, bandwidth (BW), and spreading factor (SF) are varied, and connectivity is evaluated in terms of signal-to-noise ratio (SNR) and received signal strength indicator (RSSI). From the experimental results, it has been observed that in an obstacle-free environment, SNR and RSSI depict the best values when SF is 7 or 8 and BW is 500 kHz. With obstacles such as two walls between the edge device and the gateway, the best values for SNR and RSSI are obtained at an SF of 10 or 11 with a BW of 125 kHz. The Arduino Nano 33 BLE Sense Rev2 was used to record accelerometer data for eight hand activities, and a convolutional neural network (CNN) was used to classify them with an average accuracy of 95.32%. The hand movements considered here are useful for rehabilitation applications. This work reveals that scenario-based LoRa parameter selection and edge computing improve the energy efficiency of the AIoT setup, which may be useful for hand movement classification.