Quantitative remote sensing estimation is critical for environmental monitoring, providing continuous measures of vegetation indices, canopy height, and carbon stock. Traditional radiative-transfer models and empirical regressions require expert knowledge and generalize poorly, while deep learning methods remain task-specific. We propose SatelliteCalculator+, a DINOv3-powered multi-task foundation model for continuous regression of spectral and structural variables. The framework combines prompt-driven cross-attentive adapters with lightweight MLP decoders, enabling efficient dense prediction from frozen features. To overcome limited supervision, we synthesize over one million paired samples from SPOT 6/7 imagery using physically defined formulas. On the Open-Canopy dataset, SatelliteCalculator+ achieves competitive accuracy across eight ecological variables while reducing inference cost, demonstrating the promise of self-supervised transformers and scalable multi-task learning for large-scale Earth observation.

Adversarial learning has achieved great success for unsupervised domain adaptation (UDA). Existing adversarial UDA methods leverage the predicted discriminative information with Nuclear-norm Wasserstein discrepancy for feature alignment. However, the limited memory space makes it very difficult to accurately calculate the Nuclear-norm, which hinders domain adaptation. To address this challenge, we propose a multi-batch Nuclear-norm adversarial network, termed as MBAN. Specifically, we build a dynamic queue to cache features, which encourages to generate a large and consistent output matrix, enabling accurate calculation of the Nuclear-norm. Then, the multi-batch Nuclear-norm discrepancy is proposed, which can effectively improve the transferability and discriminability of the learned features. Experimental results show that MBAN could achieve significant performance improvement, especially when the number of categories is quite large. Code is available at https://github.com/peiwang0518/Multi-BAN.

Adversarial domain adaptation has achieved impressive performances for unsupervised domain adaptation (UDA). However, existing adversarial UDA methods often rely on multiple domain discriminators to capture diverse patterns, which limit their scalability and resulting in dispersed features. To address these issues, we propose a novel method called Class-aware Prototypical Adversarial Network (CaPAN), which efficiently extracts transferable and discriminative features. Specifically, our class-aware adversarial learning employs a single multi-class discriminator to capture various patterns, aligning class-level features. Furthermore, to enhance the discriminative ability of our model, we introduce a prototypical domain discriminator to enhance the discriminatively of the learned features by aligning target sample towards prototypes (centers of each class), resulting in a more compact feature space. Extensive experiments validate the effectiveness of CaPAN, which can also be integrated as a regularization technique for existing methods to further improve their performance. Code is available at https://github.com/YuZhenyuLindy/CaPAN.

Unsupervised domain adaptation, which transfers knowledge from the source domain to the target domain, has still been a challenging problem. However, previous domain adaptation methods typically minimize the domain discrepancy by using the pseudo target labels. Since the pseudo labels can be noisy, which may cause misalignment and unsatisfying adaptation performance. To address the above challenges, we propose an information maximization adaptation network with label distribution priors. We revisit feature alignment in unsupervised domain adaptation from the perspective of distribution alignment, and find that learning discriminant feature representation requires to minimizing distribution discrepancy and maximizing source mutual information between the outputs of the classifier and feature representations. Due to domain shift, maximizing target mutual information may align features to incorrect class directly. We propose a weighted target mutual information by re-weighting the estimated mutual information via the mean prediction confidence in mini-batch, which can eliminate the negative impact of inaccurate estimation. In addition, we introduce a regularization term of label priors distribution to encourage the similarity to the real label distribution. Extensive experimental results on three benchmark datasets show that our proposed method can achieve remarkable results compared with previous methods.