AgroLens: A low-cost and green-friendly Smart Farm Architecture to support real-time leaf disease diagnostics

Abstract

Agriculture is one of the most significant global economic activities responsible for feeding the world population of 7.75 billion. However, weather conditions and diseases impact production efficiency, reducing economic activity and the food sovereignty of economies worldwide. Thus, computational methods can support disease classification based on an image. This classification requires training Artificial Intelligence (AI) models on high-performance computing resources, usually far from the user domain. State of the art has proposed the concept of Edge Computing (EC), which aims to bring computational resources closer to the domain problem to decrease application latency and improve computational power closer to the client. In addition, EC has become an enabling technology for Smart Farms, and the literature has appropriated EC to support these applications. However, predominantly state-of-the-art architectures are dependent on Internet connectivity and do not allow diverse real-time classification of diseases based on crop leaf on mobile devices. This paper sheds light on a new architecture, AgroLens, built with low-cost and green-friendly devices to support a mobile Smart Farm application, operational even in areas lacking Internet connectivity. Among our main contributions, we highlight the functional evaluation of AgroLens for AI-based real-time classification of diseases based on leaf images, achieving high classification performance using a smartphone. Our results indicate that AgroLens supports the connectivity of thousands of sensors from a smart farm without imposing computational overhead on edge-compute. The AgroLens architecture opens up opportunities and research avenues for deployment and evaluation for large-scale Smart Farm applications with low-cost devices.