Enhancing Smart Retail with Lightweight Deep Learning and Scalable Edge Inference

In these days, smart retail systems have grown to offer seamless and automated shopping experiences by utilizing Artificial Intelligence (AI) and edge computing. However, traditional architectures faced challenges with identification of real-time product and scalability across numerous shelves. To address these challenges, this research proposes a novel Lightweight Edge AI-Enabled Smart Retail System (LEAI-SRS) framework. Initially, the data is captured by a camera-enabled Internet of Things (IoT) shelves and processed locally on Raspberry Pi devices clustered under a unified architecture. Moreover, the system uses MobileNet Deep Learning (DL) model, which allows fast and accurate product recognition on constrained hardware. Furthermore, image preprocessing techniques are employed to improve visual clarity under variable lighting states, enhancing recognition stability. Furthermore, for larger setups, inference is offloaded to an edge AI server like NVIDIA Jetson Nano, to support scalable multi-shelf configurations. Additionally, the product details are modified in real time via firebase cloud, which automatically manages stock and billing without human interaction. The proposed LEAI-SRS system attained better results in terms of inference time (750 ms) and accuracy (97.2%) which outperforming existing Raspberry Pi-based systems.