Design and Implementation of a Lightweight AI-Driven Real-Time Preprocessing System for Remote Sensing Imagery

Abstract

Thanks to the intensive development of remote sensing systems, Earth observation has never been accessed at such a scale, and thus, the urgency has been revealed to develop an effective and timely preprocessing response. Conventional preprocessing techniques, mostly operating on physical models, are frequently computationally expensive, fixed, and cannot be implemented in resource-constrained systems, such as satellites, unmanned aerial vehicles and edge devices. Recent technological progress in artificial intelligence, and especially lightweight deep learning models, presents a potential alternative since it provides adaptive, data-driven preprocessing at the cost of less computation. This review provides an in-depth discussion of the design and implementation of lightweight AI-based real-time preprocessing platforms for remote sensing imagery. It analyzes the basic properties of remote sensing data, important preprocessing activities, and how model-driven and data-driven approaches can be changed to each other. The paper goes on to discuss lightweight AI methods, such as model compression and efficient architecture, and system-level design considerations, such as hardware platforms, edge and cloud integration, and pipeline optimization. Practical challenges and opportunities are presented with references to the real-life situations in disaster monitoring, precision agriculture, and urban analytics. This work is a novel one due to its holistic approach, which focuses on the introduction of the innovation of algorithms and the deployment of systems. Lastly, future directions are mentioned to help formulate scalable, efficient, and robust preprocessing structures.