A Comparison of Quantization Techniques for Mobile Olive Disease Detection

Abstract

This study aims to examines the integration of artificial intelligence (AI) in edge devices, addressing challenges such as memory, energy, and computational constraints through model quantization techniques. It focuses on Post-Training Quantization (PTQ), which simplifies the process by applying quantization after training, and Quantization-Aware Training (QAT), which incorporates quantization during training to enhance accuracy. A comparative analysis highlights that while both methods reduce model size and improve latency, QAT generally preserves higher accuracy. The practical application of these optimized models is demonstrated in a mobile application for identification of olive diseases using embedded intelligent as a study case, showcasing the benefits and challenges of deploying AI on edge devices.