Lightweight Image Segmentation for Sustainable Smart Farming using Cloud–Edge AI Collaboration

Precision agriculture requires accurate, real-time crop and disease segmentation to enable site-specific management with minimal resource use. This paper consolidates our proposed Cloud–Edge collaborative approach for lightweight semantic segmentation on resource-constrained devices. We combine classical image pre-processing (median filtering and resizing) with a compact deep model tailored for edge deployment and evaluate the pipeline on four horticultural crops (mango, sweet orange, chilli, and tomato). We demonstrate feasibility on a Raspberry Pi edge node, present qualitative results, confidence/precision–recall curves, and a normalized confusion matrix extracted from the experimental dashboard. The study highlights the design choices that favor energy- and memory-efficient inference without cloud dependence, and discusses remaining challenges in generalization, dataset diversity, and multispectral sensing.