Chronological Review and Performance Analysis of YOLO-Based Deep Learning Frameworks in Complex Geospatial Environments

The rapid evolution of YOLO (You Only Look Once) models has transformed object detection into a real-time, scalable solution across a variety of domains. This article presents a comprehensive and chronological review of YOLO-based frameworks, emphasizing their adoption in remote sensing applications. It outlines foundational concepts in object detection, including detection paradigms, evaluation metrics, and training strategies, before analyzing each YOLO version from v1 through v12, along with key architectural innovations and variants such as YOLOX, YOLOR, and DAMO-YOLO. A particular focus is placed on geospatial use cases which are urban infrastructure mapping, disaster monitoring, agricultural analysis, and security surveillance, where these models demonstrate exceptional potential. The study also examines current datasets, performance benchmarking across platforms, and post-processing pipelines. Critical limitations such as small object detection, domain shift, and class imbalance are discussed, followed by an exploration of emerging directions involving transformers, foundation models, multimodal fusion, and context-aware detection. This survey serves as a detailed resource for researchers seeking to develop, deploy, or extend YOLO-based models in remote sensing and beyond.