A TinyML-based system for gas leakage detection

Internet of Things devices are commonly utilized in smart homes to provide smart services such as lighting, entertainment, and easy access, but they are also employed to warn occupants in the event of an emergency. Due to the computationally intensive nature of existing Neural Network implementations, data must be transmitted to the cloud for analysis to produce intelligent machines. TinyML is a promising approach that the scientific community has proposed as a method of constructing autonomous and secure devices that can collect, analyze, and output data without requiring it to be transferred to remote entities. This work presents a TinyML-based system for detecting hazardous gas leaks. The system may be trained to detect irregularities and notify occupants using BLE technology via a message sent to their smartphones, as well as through an integrated screen. Two different test cases are presented and evaluated via experiments. For the smoke detection test case the system achieved an F1-Score of 0.77, whereas for the ammonia test case the evaluation metric of F1-Score is 0.70.