Mapping deforestation is an essential step in the process of managing tropical rainforests. It lets us understand and monitor both legal and illegal deforestation and its implications, which include the effect deforestation may have on climate change through greenhouse gas emissions. Given that there is ample room for improvements when it comes to mapping deforestation using satellite imagery, in this study, we aimed to test and evaluate the use of algorithms belonging to the growing field of deep learning (DL), particularly convolutional neural networks (CNNs), to this end. Although studies have been using DL algorithms for a variety of remote sensing tasks for the past few years, they are still relatively unexplored for deforestation mapping. We attempted to map the deforestation between images approximately one year apart, specifically between 2017 and 2018 and between 2018 and 2019. Three CNN architectures that are available in the literature—SharpMask, U-Net, and ResUnet—were used to classify the change between years and were then compared to two classic machine learning (ML) algorithms—random forest (RF) and multilayer perceptron (MLP)—as points of reference. After validation, we found that the DL models were better in most performance metrics including the Kappa index, F1 score, and mean intersection over union (mIoU) measure, while the ResUnet model achieved the best overall results with a value of 0.94 in all three measures in both time sequences. Visually, the DL models also provided classifications with better defined deforestation patches and did not need any sort of post-processing to remove noise, unlike the ML models, which needed some noise removal to improve results.

The Synthetic Aperture Radar (SAR) time series allows describing the rice phenological cycle by the backscattering time signature. Therefore, the advent of the Copernicus Sentinel-1 program expands studies of radar data (C-band) for rice monitoring at regional scales, due to the high temporal resolution and free data distribution. Recurrent Neural Network (RNN) model has reached state-of-the-art in the pattern recognition of time-sequenced data, obtaining a significant advantage at crop classification on the remote sensing images. One of the most used approaches in the RNN model is the Long Short-Term Memory (LSTM) model and its improvements, such as Bidirectional LSTM (Bi-LSTM). Bi-LSTM models are more effective as their output depends on the previous and the next segment, in contrast to the unidirectional LSTM models. The present research aims to map rice crops from Sentinel-1 time series (band C) using LSTM and Bi-LSTM models in West Rio Grande do Sul (Brazil). We compared the results with traditional Machine Learning techniques: Support Vector Machines (SVM), Random Forest (RF), k-Nearest Neighbors (k-NN), and Normal Bayes (NB). The developed methodology can be subdivided into the following steps: (a) acquisition of the Sentinel time series over two years; (b) data pre-processing and minimizing noise from 3D spatial-temporal filters and smoothing with Savitzky-Golay filter; (c) time series classification procedures; (d) accuracy analysis and comparison among the methods. The results show high overall accuracy and Kappa (>97% for all methods and metrics). Bi-LSTM was the best model, presenting statistical differences in the McNemar test with a significance of 0.05. However, LSTM and Traditional Machine Learning models also achieved high accuracy values. The study establishes an adequate methodology for mapping the rice crops in West Rio Grande do Sul.

The need to monitor the Earth’s surface over a range of spatial and temporal scales is fundamental in ecosystems planning and management. Change-Vector Analysis (CVA) is a bi-temporal method of change detection that considers the magnitude and direction of change vector. However, many multispectral applications do not make use of the direction component. The procedure most used to calculate the direction component using multiband data is the direction cosine, but the number of output direction cosine images is equal to the number of original bands and has a complex interpretation. This paper proposes a new approach to calculate the spectral direction of change, using the Spectral Angle Mapper and Spectral Correlation Mapper spectral-similarity measures. The chief advantage of this approach is that it generates a single image of change information insensitive to illumination variation. In this paper the magnitude component of the spectral similarity was calculated in two ways: as the standard Euclidean distance and as the Mahalanobis distance. In this test the best magnitude measure was the Euclidean distance and the best similarity measure was Spectral Angle Mapper. The results show that the distance and similarity measures are complementary and need to be applied together.

Remote sensing has been used in karst studies to identify limestone terrain, describe exokarst features, analyze karst depressions, and detect geological structures important to karst development. The aim of this work is to investigate the use of ASTER-, SRTM- and ALOS/PRISM-derived digital elevation models (DEMs) to detect and quantify natural karst depressions along the São Francisco River near Barreiras city, northeast Brazil. The study area is a karst landscape characterized by karst depressions (dolines), closed depressions in limestone, many of which contain standing water connected with the ground-water table. The base of dolines is typically sealed with an impermeable clay layer covered by standing water or herbaceous vegetation. We identify dolines by combining the extraction of sink depth from DEMs, morphometric analysis using GIS, and visual interpretation. Our methodology is a semi-automatic approach involving several steps: (a) DEM acquisition; (b) sink-depth calculation using the difference between the raw DEM and the corresponding DEM with sinks filled; and (c) elimination of falsely identified karst depressions using morphometric attributes. The advantages and limitations of the applied methodology using different DEMs are examined by comparison with a sinkhole map generated from traditional geomorphological investigations based on visual interpretation of the high-resolution remote sensing images and field surveys. The threshold values of the depth, area size and circularity index appropriate for distinguishing dolines were identified from the maximum overall accuracy obtained by comparison with a true doline map. Our results indicate that the best performance of the proposed methodology for meso-scale karst feature detection was using ALOS/PRISM data with a threshold depth > 2 m; areas > 13,125 m² and circularity indexes > 0.3 (overall accuracy of 0.53). The overall correct identification of around half of the true dolines suggests the potential to substantially improve doline identification using higher-resolution LiDAR-generated DEMs.

Instance segmentation is the state-of-the-art in object detection, and there are numerous applications in remote sensing data where these algorithms can produce significant results. Nevertheless, one of the main problems is that most algorithms use Red, Green, and Blue (RGB) images, whereas Satellite images often present more channels that can be crucial to improve performance. Therefore, the present work brings three contributions: (a) conversion system from ground truth polygon data into the Creating Common Object in Context (COCO) annotation format; (b) Detectron2 software source code adaptation and application on multi-channel imagery; and (c) large scene image mosaicking. We applied the procedure in a Center Pivot Irrigation System (CPIS) dataset with ground truth produced by the Brazilian National Water Agency (ANA) and Landsat-8 Operational Land Imager (OLI) imagery (7 channels with 30-m resolution). Center pivots are a modern irrigation system technique with massive growth potential in Brazil and other world areas. The round shapes with different textures, colors, and spectral behaviors make it appropriate to use Deep Learning instance segmentation. We trained the model using 512 × 512-pixel sized patches using seven different backbone structures (ResNet50- Feature Pyramid Network (FPN), Resnet50-DC5, ResNet50-C4, Resnet101-FPN, Resnet101-DC5, ResNet101-FPN, and ResNeXt101-FPN). The model evaluation used standard COCO metrics (Average Precision (AP), AP₅₀, AP₇₅, AP_small, AP_medium, and AR₁₀₀). ResNeXt101-FPN had the best results, with a 3% advantage over the second-best model (ResNet101-FPN). We also compared the ResNeXt101-FPN model in the seven-channel and RGB imagery, where the multi-channel model had a 3% advantage, demonstrating great improvement using a larger number of channels. This research is also the first with a mosaicking algorithm using instance segmentation models, where we tested in a 1536 × 1536-pixel image using a non-max suppression sorted by area method. The proposed methodology is innovative and suitable for many other remote sensing problems and medical imagery that often present more channels.