Accurate crop type identification and crop area estimation from remote sensing data in tropical regions are still considered challenging tasks. The more favorable weather conditions, in comparison to the characteristic conditions of temperate regions, permit higher flexibility in land use, planning, and management, which implies complex crop dynamics. Moreover, the frequent cloud cover prevents the use of optical data during large periods of the year, making SAR data an attractive alternative for crop mapping in tropical regions. This paper evaluates the effectiveness of Deep Learning (DL) techniques for crop recognition from multi-date SAR images from tropical regions. Three DL strategies are investigated: autoencoders, convolutional neural networks, and fully-convolutional networks. The paper further proposes a post-classification technique to enforce prior knowledge about crop dynamics in the target area. Experiments conducted on a Sentinel-1 multitemporal sequence of a tropical region in Brazil reveal the pros and cons of the tested methods. In our experiments, the proposed crop dynamics model was able to correct up to 16.5% of classification errors and managed to improve the performance up to 3.2% and 8.7% in terms of overall accuracy and average F1-score, respectively.

The Amazon rainforest spreads across nine countries and covers nearly one-third of South America, being 69% inside Brazilian borders. It represents more than half of the remaining tropical forest on Earth and covers the catchment basin of the Amazon river on which 20% of the surface fresh water on the planet flows. Such an ecosystem produces large quantities of water vapor, helping regulate rainfall regimes in most of South America, with strong economic implications: for instance, by irrigating crops and pastures, and supplying water for the main hydroelectric plants in the continent. Being the natural habitat of one-tenth of the currently known species, the Amazon also has enormous biotechnological potential. Among the major menaces to the Amazon is the extension of agricultural and cattle farming, forest fires, illegal mining and logging, all directly associated with deforestation. Preserving the Amazon is obviously essential, and it is well-known that remote sensing provides effective tools for environmental monitoring. This work presents a deforestation detection approach based on the DeepLabv3+, a fully convolutional deep learning model devised for semantic segmentation. The proposed method extends the original DeepLabv3+ model, aiming at properly dealing with a strong class imbalanced problem and improving the delineation quality of deforestation polygons. Experiments were devised to evaluate the proposed method in terms of the sensitivity to the weighted focal loss hyperparameters—through an extensive grid search—and the amount of training data, and compared its performance to previous deep learning methods proposed for deforestation detection. Landsat OLI-8 images of a specific region in the Amazon were used in such evaluation. The results indicate that the variants of the proposed method outperformed previous works in terms of the F1-score and Precision metrics. Additionally, more substantial performance gains were observed in the context of smaller volumes of training data. When the evaluated methods were trained using four image tiles, the proposed method outperformed its counterparts by approximately +10% in terms of F1-score (from 63% to 73%); when the methods were trained with only one image tile, the performance difference in terms of F1-score achieved approximately +18% (from 49% to 67%).

Many deep-learning (DL)-based, domain adaptation (DA) methods for remote sensing (RS) applications rely on adversarial training strategies to align features extracted from images of different domains in a shared latent space. However, the performance of such representation matching techniques is negatively impacted when class occurrences in the target domain, for which no labeled data are available during training, are highly imbalanced. In this work, we propose a DL-based representation matching approach for DA in the context of change detection tasks. We further evaluate the approach in a deforestation mapping application, characterized by a high-class imbalance between the deforestation and no-deforestation classes. The domains represent different sites in the Amazon and Brazilian Cerrado biomes. To mitigate the class imbalance problem, we devised an unsupervised pseudolabeling scheme based on change vector analysis (CVA) that prevents the feature alignment to be biased toward the overrepresented class. The experimental results indicate that the proposed approach can improve the accuracy of cross-domain deforestation detection.

In this work, we introduce a novel, distributed version of the N-FINDR endmember extraction algorithm, which is able to exploit computer cluster resources in order to efficiently process large volumes of hyperspectral data. The implementation of the distributed algorithm was done by extending the InterCloud Data Mining Package, originally adopted for land cover classification, through the HyperCloud-RS framework, here adapted for endmember extraction, which can be executed on cloud computing environments, allowing users to elastically administer processing power and storage space for adequately handling very large datasets. The framework supports distributed execution, network communication, and fault tolerance, transparently and efficiently to the user. The experimental analysis addresses the performance issues, evaluating both accuracy and execution time, over the processing of different synthetic versions of the AVIRIS Cuprite hyperspectral dataset, with 3.1 Gb, 6.2 Gb, and 15.1Gb respectively, thus addressing the issue of dealing with large-scale hyperspectral data. As a further contribution of this work, we describe in detail how to extend the HyperCloud-RS framework by integrating other endmember extraction algorithms, thus enabling researchers to implement algorithms specifically designed for their own assessment.

Domain adaptation has proven to be suitable for alleviating domain discrepancies, which hinder the generalization capacity of classifiers. Among a few alternatives, domain adaptation techniques that align features in a domain-agnostic space through adversarial learning have been widely investigated. Nevertheless, such an approach often implies the deterioration of feature discriminability as a side effect of the adversarial alignment, which does not take into consideration class labels of the target domain samples. We advocate that weakly-supervised learning can help mitigate that problem, as noisy labels for the target domain samples may serve to sustain class discriminability during the feature alignment procedure. Therefore, in this work we propose a weakly-supervised, adversarial domain adaptation method for a change detection task based on the Domain Adversarial Neural Network (DANN) strategy. We assessed the performance of the proposed method on a deforestation detection application, conducting experiments in sites of the Amazon and Cerrado (Brazilian Savannah) biomes using Landsat-8 images, each site corresponding to a different domain. The results showed that the inclusion of weak supervision in the domain adaptation procedure provided higher accuracies as compared to the original DANN strategy, which did not prescribe any supervision for the selection of target domain samples in the training procedure. On average, the Average Precision and F1-score metrics values increased by 10.1 use of the proposed method. Additionally, our method achieved performances that are compatible with the ones obtained with previously proposed, state-of-the-art domain adaptation methods. To the best of our knowledge, the proposed method is the first weakly-supervised domain adaptation strategy conceived for deforestation detection and, in general, for change detection.