Search

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 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 on sites of the Amazon and Cerrado biomes using Landsat-8 images. The results showed that the inclusion of weak supervision in the domain adaptation procedure provided higher accuracies than the original DANN strategy, which did not prescribe any supervision for the selection of target domain samples in training. On average, the Average Precision and F1-score values increased by 10.1\% and 12.6\% respectively with the use of the proposed method. Additionally, our method achieved compatible performances with the ones obtained by 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.

This paper presents an algorithm based on Convolutional Neural Networks (CNN) to find the depth and angles of inclination and rotation of a foreign object inside the human body based on images of the magnetic field generated by it. The key challenge is to provide information with enough accuracy to be used in surgical procedures. We tested three distinct CNN architectures for values prediction, and our best model achieved a mean - average F1-score of 66 %, 100 %, and 98 % in the test dataset for depth and angles of inclination and rotation, respectively. We also propose an approach for converting classification values to real values and we calculate the type A uncertainty for all models, with our best model showing an uncertainty of 10.8 mm for depth, 2.0° for inclination and 7.5° for rotation values.

Detecting early deforestation is a fundamental process in reducing forest degradation and carbon emissions. With this procedure, it is possible to monitor and control illegal activities associated with deforestation. Most regular monitoring projects have been recently proposed, but most of them rely on optical imagery. In addition, these data are seriously restricted by cloud coverage, especially in tropical environments. In this regard, Synthetic Aperture Radar (SAR) is an attractive alternative that can fill this observational gap. This work evaluated and compared a conventional method based on time series and a Fully Convolutional Network (FCN) with bi-temporal SAR images. These approaches were assessed in two regions of the Brazilian Amazon to detect deforestation between 2019 and 2020. Different pre-processing techniques, including filtering and stabilization stages, were applied to the C-band Sentinel-1 images. Furthermore, this study proposes to provide the network with the distance map to past-deforestation as additional information to the pair of images being compared. In our experiments, this proposal brought up to 4% improvement in average precision. The experimental results further indicated a clear superiority of the DL approach over a time series-based deforestation detection method used as a baseline in all experiments. Finally, the study proved the benefits of pre-processing techniques when using detection methods based on time series. On the contrary, the analysis revealed that the neural network could eliminate noise from the input images, making filtering innocuous and, therefore, unnecessary. On the other hand, the stabilization of the input images brought non-negligible accuracy gains to the DL approach.

The availability of remote-sensing multisource data from optical-based satellite sensors has created new opportunities and challenges for forest monitoring in the Amazon Biome. In particular, change-detection analysis has emerged in recent decades to monitor forest-change dynamics, supporting some Brazilian governmental initiatives such as PRODES and DETER projects for biodiversity preservation in threatened areas. In recent years fully convolutional network architectures have witnessed numerous proposals adapted for the change-detection task. This paper comprehensively explores state-of-the-art fully convolutional networks such as U-Net, ResU-Net, SegNet, FC-DenseNet, and two DeepLabv3+ variants on monitoring deforestation in the Brazilian Amazon. The networks’ performance is evaluated experimentally in terms of Precision, Recall, F1-score, and computational load using satellite images with different spatial and spectral resolution: Landsat-8 and Sentinel-2. We also include the results of an unprecedented auditing process performed by senior specialists to visually evaluate each deforestation polygon derived from the network with the highest accuracy results for both satellites. This assessment allowed estimation of the accuracy of these networks simulating a process “in nature” and faithful to the PRODES methodology. We conclude that the high resolution of Sentinel-2 images improves the segmentation of deforestation polygons both quantitatively (in terms of F1-score) and qualitatively. Moreover, the study also points to the potential of the operational use of Deep Learning (DL) mapping as products to be consumed in PRODES.

Per-point classification is a traditional method for remote sensing data classification, and for radar data in particular. Compared with optical data, the discriminative power of radar data is quite limited, for most applications. A way of trying to overcome these difficulties is to use Region-Based Classification (RBC), also referred to as Geographical Object-Based Image Analysis (GEOBIA). RBC methods first aggregate pixels into homogeneous objects, or regions, using a segmentation procedure. Moreover, segmentation is known to be an ill-conditioned problem because it admits multiple solutions, and a small change in the input image, or segmentation parameters, may lead to significant changes in the image partitioning. In this context, this paper proposes and evaluates novel approaches for SAR data classification, which rely on specialized segmentations, and on the combination of partial maps produced by classification ensembles. Such approaches comprise a meta-methodology, in the sense that they are independent from segmentation and classification algorithms, and optimization procedures. Results are shown that improve the classification accuracy from Kappa = 0.4 (baseline method) to a Kappa = 0.77 with the presented method. Another test site presented an improvement from Kappa = 0.36 to a maximum of 0.66 also with radar data.

Deforestation is one of the major threats to natural ecosystems. This process has a substantial contribution to climate change and biodiversity reduction. Therefore, the monitoring and early detection of deforestation is an essential process for preservation. Techniques based on satellite images are among the most attractive options for this application. However, many approaches involve some human intervention or are dependent on a manually selected threshold to identify regions that suffer deforestation. Motivated by this scenario, the present work evaluates Deep Learning-based strategies for automatic deforestation detection, namely, Early Fusion (EF), Siamese Network (SN), and Convolutional Support Vector Machine (CSVM) as well as Support Vector Machine (SVM), used as the baseline. The target areas are two regions with different deforestation patterns: the Amazon and Cerrado biomes in Brazil. The experiments used two co-registered Landsat 8 images acquired at different dates. The strategies based on Deep Learning achieved the best performance in our analysis in comparison with the baseline, with SN and EF superior to CSVM and SVM. In the same way, a reduction of the salt-and-pepper effect in the generated probabilistic change maps was noticed as the number of training samples increased. Finally, the work assesses how the methods can reduce the time invested in the visual inspection of deforested areas.

This study proposes and evaluates five deep fully convolutional networks (FCNs) for the semantic segmentation of a single tree species: SegNet, U-Net, FC-DenseNet, and two DeepLabv3+ variants. The performance of the FCN designs is evaluated experimentally in terms of classification accuracy and computational load. We also verify the benefits of fully connected conditional random fields (CRFs) as a post-processing step to improve the segmentation maps. The analysis is conducted on a set of images captured by an RGB camera aboard a UAV flying over an urban area. The dataset also contains a mask that indicates the occurrence of an endangered species called Dipteryx alata Vogel, also known as cumbaru, taken as the species to be identified. The experimental analysis shows the effectiveness of each design and reports average overall accuracy ranging from 88.9% to 96.7%, an F1-score between 87.0% and 96.1%, and IoU from 77.1% to 92.5%. We also realize that CRF consistently improves the performance, but at a high computational cost.

Detection of urban area extents by means of remotely sensed data is a difficult task, especially because of the multiple, diverse definitions of what an “urban area” is. The models of urban areas listed in technical literature are based on the combination of spectral information with spatial patterns, possibly at different spatial resolutions. Starting from the same data set, “urban area” extraction may thus lead to multiple outputs. If this is done in a well-structured framework, however, this may be considered as an advantage rather than an issue. This paper proposes a novel framework for urban area extent extraction from multispectral Earth Observation (EO) data. The key is to compute and combine spectral and multi-scale spatial features. By selecting the most adequate features, and combining them with proper logical rules, the approach allows matching multiple urban area models. Experimental results for different locations in Brazil and Kenya using High-Resolution (HR) data prove the usefulness and flexibility of the framework.

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.

Detection and classification of tree species from remote sensing data were performed using mainly multispectral and hyperspectral images and Light Detection And Ranging (LiDAR) data. Despite the comparatively lower cost and higher spatial resolution, few studies focused on images captured by Red-Green-Blue (RGB) sensors. Besides, the recent years have witnessed an impressive progress of deep learning methods for object detection. Motivated by this scenario, we proposed and evaluated the usage of Convolutional Neural Network (CNN)-based methods combined with Unmanned Aerial Vehicle (UAV) high spatial resolution RGB imagery for the detection of law protected tree species. Three state-of-the-art object detection methods were evaluated: Faster Region-based Convolutional Neural Network (Faster R-CNN), YOLOv3 and RetinaNet. A dataset was built to assess the selected methods, comprising 392 RBG images captured from August 2018 to February 2019, over a forested urban area in midwest Brazil. The target object is an important tree species threatened by extinction known as Dipteryx alata Vogel (Fabaceae). The experimental analysis delivered average precision around 92% with an associated processing times below 30 miliseconds.

O presente trabalho apresenta uma metodologia para modelar conhecimento multitemporal em classificação automática de imagens. Este método utiliza diagramas de transição de estado para representar as possibilidades de mudanças de classe dentro de um determinado intervalo de tempo. As possibilidades de mudança são estimadas a partir de dados previamente adquiridos da mesma região usando algoritmos genéticos. Os resultados da classificação são obtidos pela combinação dos conhecimentos multitemporal e espectral usando técnicas da lógica nebulosa. O método proposto apresentou bons resultados quando testado utilizando imagens LANDSAT 7 de três anos consecutivos de uma área no sudoeste do Brasil. O conhecimento multitemporal sempre melhorou o desempenho da classificação em comparação à classificação puramente espectral. Além disso, o algoritmo genético foi, em todos os casos, superior à alternativa manual de tentativa e erro para estimar parâmetros do modelo multitemporal.

Este trabalho dedica-se a explorar métodos cognitivos de classificação de imagens QuickBird, visando detectar áreas residenciais homogêneas. A área de estudo localiza-se em um setor sul da cidade de São José dos Campos, São Paulo. Por meio de uma abordagem baseada em segmentação multinível e de uma rede hierárquica de classificação com seis níveis, identificaram-se áreas residenciais homogêneas em termos de densidade de ocupação e de grau de aproveitamento do terreno por parte das edificações, isto é, residências unifamiliares ou edifícios. A rede de classificação utilizou atributos espectrais, geométricos e topológicos dos objetos (ou regiões) em cada nível de segmentação, bem como suas características contextuais e semânticas de relacionamento entre os distintos níveis hierárquicos. A classificação final de unidades residenciais homogêneas foi submetida a um processo de validação, utilizando-se um índice de concordância Kappa baseado em objetos.

Este artigo introduz uma nova plataforma cognitiva open source destinada à interpretação automática de imagens digitais, denominada InterIMAGE. Serão apresentados neste trabalho os aspectos principais, a arquitetura, bem como uma visão geral da estratégia de interpretação implementada no InterIMAGE. Será também apresentado um experimento, no qual uma aplicação construída e testada em uma plataforma comercial (eCognition) é transposta para o InterIMAGE. Nenhuma mudança foi introduzida no modelo original de representação do conhecimento propriamente dito, apenas a maneira de formulação do conhecimento foi adaptada, de modo a ajustar-se às estruturas de representação do conhecimento no InterIMAGE. Os resultados apresentaram classificações praticamente idênticas produzidas através de ambas as plataformas.