Cost-efficient high-resolution monitoring for specialty crops using AgI-GAN and AI-driven analytics
Christian Lacerda, Yiannis G. Ampatzidis, Antônio de Oliveira Costa Neto, Victor Partel · Computers and Electronics in Agriculture · 2025
Unmanned aerial vehicles (UAVs) and airplane or satellite imagery are widely used technologies for data acquisition in precision agriculture, each with advantages and disadvantages. UAVs capture high-resolution images but have a limited collection range, leading to higher costs per acre. In contrast, airplane imagery is more cost-effective and covers larger areas but typically provides lower spatial resolution (15–50 cm), which is inadequate for assessing specialty crops. For example, Agroview, an AI-driven platform for plant inventories and plant-level analytics, requires high-resolution imagery (<10 cm/pixel) for optimal performance. This study aims to address this limitation by developing an agriculture intelligence generative adversarial network (AgI-GANs) to enhance airplane imagery (∼15 cm/pixel resolution) to resolutions comparable to UAV imagery (∼3.9 cm/pixel). The AgI-GAN-enhanced images significantly improved Agroview’s tree detection accuracy, reducing detection errors to below 10 %, with an average error of 4.11 % ± 4.5 across 10 citrus orchards, compared to 36.21 % ± 9 when using unenhanced airplane images. Additionally, cost analysis demonstrates that AgI-GAN-enhanced airplane imagery reduces data acquisition costs to approximately $2.44 per acre, a 60 % reduction from the $6.07 per acre cost associated with UAV data collection. This study highlights the transformative potential of AgI-GAN in precision agriculture, providing a cost-effective solution that combines the extensive coverage of airplane imagery with the high resolution required for precise data analytics, benefiting specialty crop growers and agricultural stakeholders.