Density Map Modeling on the Edge: Towards Sustainable Agri-Food Supply Chains

While food waste remains a significant global climate burden, effective demand planning in retail can reduce overstocking, propagating toward sustainable Agri-Food Supply Chains (AFSCs). From here, this paper employs density map modeling with Edge Intelligence (Edge-AI) technology to identify the On-Shelf Availability (OSA) index in retailers. A lightweight and efficient variant of the Multi-Column Convolutional Neural Network (MCNN), referred to as Modified Multi-Column Convolutional Neural Network (mMCNN), is proposed for edge-based density estimation. Moreover, a novel dataset called AgriShelf is curated to serve as a benchmark for this application. Extensive tests on edge platforms, Jetson Nano and Jetson Orin Nano, show that mMCNN outperforms the original MCNN in computational efficiency, with only a nominal drop in counting accuracy. Notably, Jetson Orin Nano achieves superior performance, reaching an inference speed of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 0 1 5}$</tex> seconds. To this end, the integration of Edge-AI and density map modeling provides real-time insights that reduce food waste, optimize replenishments, and drive sustainable policy decisions across AFSCs.