Feasibility Study of an Electronic Nose as a Support Tool for Canine Agents

Abstract

This work addresses the feasibility study of an electronic nose (e-nose) as a complementary tool for canine agents in substance detection tasks. A dataset of hashish samples as the target substance and other substances as negative samples is constructed, training an artificial neural network with Keras. A mobile application with Bluetooth communication using TensorFlow Lite models is developed. Both models, one with original data and another with augmented data, achieve high accuracy rates (96% and 94.67%). Comparative tests with a specialized canine agent show that the e-nose is viable as a complementary tool in specific situations, although with limitations when facing air currents and complex environments.