GANs for Precision Image Generation in Field Crop Protection

Guangfang Xiang^*

• Pingdingshan University, Pingdingshan, China

Precision crop protection plays a crucial role in sustainable agriculture by enabling timely and targeted interventions against pests, diseases, and abiotic stressors. However, the effectiveness of data-driven models in this domain is severely limited by the scarcity of annotated images, the imbalanced distribution of rare crop conditions, and the high variability of field environments across time and space. Traditional data augmentation techniques-such as geometric transformations and color jittering-fail to capture the biological diversity and contextual complexity required for robust model generalization. Furthermore, conventional generative approaches are often detached from the underlying spatiotemporal dynamics and domainspecific agronomic knowledge. To address these challenges, we propose a novel generative framework that combines a Spatial-Temporal Adversarial Generation Network (ST-AGN) with an Adaptive Domain-Aware Synthesis Strategy (ADASS). The ST-AGN generates high-resolution, biologically plausible images conditioned on crop status, environmental attributes, and field location. ADASS dynamically selects and integrates synthetic samples based on domain coverage, task uncertainty, and feedback from real-time monitoring. This combination enables targeted simulation of rare or underrepresented agricultural scenarios and improves the generalization ability of downstream detection and prediction models. Experimental results on four public agricultural datasets demonstrate that our framework outperforms six state-of-the-art generative models across multiple evaluation metrics, including FID, Inception Score, and task-level precision and recall. The method not only enhances synthetic image quality but also significantly boosts performance in crop disease detection, stress classification, and resource optimization tasks. This research highlights the potential of domain-aware generative modeling as a scalable and adaptive solution to data scarcity in precision agriculture, paving the way for more intelligent, efficient, and sustainable crop protection systems.