Driven by abundant satellite imagery, machine learning-based approaches have
recently been promoted to generate high-resolution crop cultivation maps to
support many agricultural applications. One of the major challenges faced by
these approaches is the limited availability of ground truth labels. In the
absence of ground truth, existing work usually adopts the "direct transfer
strategy" that trains a classifier using historical labels collected from other
regions and then applies the trained model to the target region. Unfortunately,
the spectral features of crops exhibit inter-region and inter-annual
variability due to changes in soil composition, climate conditions, and crop
progress, the resultant models perform poorly on new and unseen regions or
years. This paper presents the Crop Generative Adversarial Network (CropGAN) to
address the above cross-domain issue. Our approach does not need labels from
the target domain. Instead, it learns a mapping function to transform the
spectral features of the target domain to the source domain (with labels) while
preserving their local structure. The classifier trained by the source domain
data can be directly applied to the transformed data to produce high-accuracy
early crop maps of the target domain. Comprehensive experiments across various
regions and years demonstrate the benefits and effectiveness of the proposed
approach. Compared with the widely adopted direct transfer strategy, the F1
score after applying the proposed CropGAN is improved by 13.13% - 50.98%

通过丰富的卫星图像和机器学习技术的驱动，该论文提出了一种名为 CropGAN 的作物生成对抗网络，以解决农作物分布图生成中源领域与目标领域之间的跨域问题，避免了目标领域标签的需求，通过学习一个映射函数将目标领域的光谱特征转换到源领域，从而实现了在目标领域产生高精度早期作物地图的目标。该方法在多个地区和年份的广泛实验中证明了其优势和有效性，与直接传输策略相比，应用 CropGAN 后的 F1 得分提高了 13.13% 至 50.98%。