Deep learning in agricultural land image classification: a comprehensive analysis of UAV-based crop identification methods

Crop identification and agricultural land cover classification have been transformed by the combination of deep learning techniques and Unmanned Aerial Vehicle (UAV) imagery. This comprehensive article systematically examines UAV imagery-based crop classification techniques, emphasizing contemporary advancements in deep learning and their practical applications in agricultural contexts. We analyse diverse UAV platforms, data collection protocols, and pre-processing techniques that form the foundation for effective classification systems. Our technical evaluation compares semantic segmentation networks (achieving up to 97.62% MIoU), specialised classification architectures (reaching 99.4% F1-score with transformer-based models), and hybrid multi-modal approaches across various agricultural scenarios. This paper presents performance metrics for various architectural designs, focusing on the development of neural network models that achieve an accuracy alongside computational efficiency. After a comprehensive analysis, the discussion transitions to identifying key challenges, including the demands for computational resources, constraints in model generalisation, scarcity of datasets, and concerns regarding environmental adaptability. This analysis outlines future research directions that prioritize the integration of multi-modal data fusion, standardised methodologies, the development of real-time processing capabilities, and extensive public datasets. Our findings indicate that despite significant advancements in deep learning-based UAV crop classification, substantial opportunities remain to enhance system robustness and practical applicability in agricultural contexts.