scholarly journals CNN SEMANTIC SEGMENTATION TO RETRIEVE PAST LAND COVER OUT OF HISTORICAL ORTHOIMAGES AND DSM: FIRST EXPERIMENTS

2020 ◽  
Vol V-2-2020 ◽  
pp. 1013-1019
Author(s):  
A. Le Bris ◽  
S. Giordano ◽  
C. Mallet
Keyword(s):  
Land Cover ◽  
Spatial Resolution ◽  
Time Series Data ◽  
Remote Sensing Data ◽  
Semantic Segmentation ◽  
Training Data ◽  
Training Dataset ◽  
Series Data ◽  
Temporal Sampling ◽  
3D Information

Abstract. Images from archival aerial photogrammetric surveys are a unique and relatively unexplored means to chronicle 3D land-cover changes occurred since the mid 20th century. They provide a relatively dense temporal sampling of the territories with a very high spatial resolution. Thus, they offer time series data which can answer a large variety of long-term environmental monitoring studies. Besides, they are generally stereoscopic surveys, making it possible to derive 3D information (Digital Surface Models). In recent years, they have often been digitized, making them more suitable to be considered in automatic analyses processes. Some photogrammetric softwares make it possible to retrieve their geometry (pose and camera calibration) and to generate corresponding DSM and orthophotomosaic. Thus, archival aerial photogrammetric surveys appear as being a powerful remote sensing data source to study land use/cover evolution over the last century. However, several difficulties have to be faced to be able to use them in automatic analysis processes. Indeed, surveys available on a study area can exhibit very different characteristics: survey pattern, focal, spatial resolution, modality (panchromatic, colour, infrared…). Planimetric and altimetric accuracies of derived products strongly depend on these characteristics. Thus, analysis processes have to cope with these uncertainties. Another important gap states in the lack of training data. Deep learning methods and especially Convolutional Neural Networks (CNN) are at present the most efficient semantic segmentation methods as long as a sufficient training dataset is available. However, temporal gaps can be very important between existing available databases and archival data. In this study, two custom variants of simple yet effective U-net - Deconv-Net inspired DL architectures are developed to process ortho-image and DSM based information. They are then trained out of a groundtruth derived out of a recent database to process archival datasets.

scholarly journals Time-Series Model-Adjusted Percentile Features: Improved Percentile Features for Land-Cover Classification Based on Landsat Data

2020 ◽  
Vol 12 (18) ◽  
pp. 3091
Author(s):  
Shuai Xie ◽  
Liangyun Liu ◽  
Jiangning Yang
Keyword(s):  
Time Series ◽  
Land Cover ◽  
Time Series Data ◽  
Land Cover Classification ◽  
Time Series Model ◽  
Training Data ◽  
Series Data ◽  
Landsat Data ◽  
Data Sampling ◽  
Continuous Change

Percentile features derived from Landsat time-series data are widely adopted in land-cover classification. However, the temporal distribution of Landsat valid observations is highly uneven across different pixels due to the gaps resulting from clouds, cloud shadows, snow, and the scan line corrector (SLC)-off problem. In addition, when applying percentile features, land-cover change in time-series data is usually not considered. In this paper, an improved percentile called the time-series model (TSM)-adjusted percentile is proposed for land-cover classification based on Landsat data. The Landsat data were first modeled using three different time-series models, and the land-cover changes were continuously monitored using the continuous change detection (CCD) algorithm. The TSM-adjusted percentiles for stable pixels were then derived from the synthetic time-series data without gaps. Finally, the TSM-adjusted percentiles were used for generating supervised random forest classifications. The proposed methods were implemented on Landsat time-series data of three study areas. The classification results were compared with those obtained using the original percentiles derived from the original time-series data with gaps. The results show that the land-cover classifications obtained using the proposed TSM-adjusted percentiles have significantly higher overall accuracies than those obtained using the original percentiles. The proposed method was more effective for forest types with obvious phenological characteristics and with fewer valid observations. In addition, it was also robust to the training data sampling strategy. Overall, the methods proposed in this work can provide accurate characterization of land cover and improve the overall classification accuracy based on such metrics. The findings are promising for percentile-based land cover classification using Landsat time series data, especially in the areas with frequent cloud coverage.

Multi-temporal mapping of pantropical post-loss land cover using dense earth-observation time series and global pre-existing maps

2020 ◽  
Author(s):  
Alejandro Coca Castro ◽  
Louis Reymondin ◽  
Mark Mulligan
Keyword(s):  
Neural Network ◽  
Time Series ◽  
Land Cover ◽  
Spatial Resolution ◽  
Time Series Data ◽  
Observation Time ◽  
Earth Observation ◽  
Series Data ◽  
Forest Change ◽  
Multi Temporal

<p>Deforestation remains one of the largest contributors to global greenhouse emissions. Despite the efforts in monitoring forest change, there is still a lack of pan-tropical spatially-explicit data informing the subsequent land cover (LC) changes over deforested areas (also known as post-loss LC). Based on this premise, this research focuses on predicting post-loss LC over deforested areas as detected by Terra-i, an early warning system of pantropical forest change providing alerts every 16-days from 2004 to present at spatial resolution of 250 m. A supervised deep neural network model suited to extract spatio-temporal patterns from dense earth observation time series data was leveraged in this work by using 16-day MODIS images of 2015. The model was trained according to nine labelled datasets representing different number of LC classes and complexity. These datasets were generated from pre-existing global LC maps with a native spatial resolution ranging from 100 m to 500 m. The effectiveness of the trained models in producing accurate predictions of post-loss LC was assessed over the Amazon region, the largest continuous region of tropical forest in the world. A two-stage assessment approach was conducted to determine the most suitable labelled datasets to predict post-loss LC over Terra-i’s areas. For the first stage, traditional metrics for the assessment of the quality of LC thematic data — e.g. overall accuracy, per-class mapping accuracy, area (or quantity) disagreement and allocation disagreement — were computed according to the test partitions from the labelled datasets. A second stage consisted in using the trained models in 2015 to make predictions for all available years of MODIS satellite imagery, from 2001 to 2018, across seven representative areas distributed in the Amazon. The observed LC predictions were masked using annual aggregated data of Terra-i from 2004 to 2010. The post-LC data by trained model, which represents a given labelled dataset, was verified by i) visualising the temporal and spatial distribution of the most frequent subsequent LC changes; and ii) comparing with Mapbiomas Amazonia, a regional-tuned multi-temporal LC dataset from 2000 to 2017 for the whole Amazon. The results showed that one out of the nine labelled datasets allowed the supervised deep learning model to produce reasonable spatial predictions and classification accuracies (overall accuracy of 86.36±0.64, area disagreement of 5.34±0.39 and allocation disagreement of 8.31±0.64) according to the test partition data. Moreover, the trained model provided similar patterns of post-loss LC as informed by the Mapbiomas dataset. Due to the nature of the model (i.e. neural network) and input data (i.e. global), it is expected the model is scalable to other pantropical areas. The insights and products derived throughout this study are targeted to reduce current uncertainties and challenges in the calculation of global and regional drivers and impacts of deforestation in tropical forests and landscapes.</p>

scholarly journals SEMANTIC SEGMENTATION OF BURNED AREAS IN SATELLITE IMAGES USING A U-NET-BASED CONVOLUTIONAL NEURAL NETWORK

2021 ◽  
Vol XLIII-B3-2021 ◽  
pp. 47-53
Author(s):  
A. K. Brand ◽  
A. Manandhar
Keyword(s):  
Central Africa ◽  
Remote Sensing Data ◽  
Semantic Segmentation ◽  
Training Data ◽  
Local Model ◽  
Training Dataset ◽  
Burned Area ◽  
Burned Areas ◽  
Research Areas ◽  
Rule Sets

Abstract. The use of remote sensing data for burned area mapping hast led to unprecedented advances within the field in recent years. Although threshold and traditional machine learning based methods have successfully been applied to the task, they implicate drawbacks including the involvement of complex rule sets and requirement of previous feature engineering. In contrast, deep learning offers an end-to-end solution for image analysis and semantic segmentation. In this study, a variation of U-Net is investigated for mapping burned areas in mono-temporal Sentinel-2 imagery. The experimental setup is divided into two phases. The first one includes a performance evaluation based on test data, while the second serves as a use case simulation and spatial evaluation of training data quality. The former is especially designed to compare the results between two local (trained only with data from the respective research areas) and a global (trained with the whole dataset) variant of the model with research areas being Indonesia and Central Africa. The networks are trained from scratch with a manually generated customized training dataset. The application of the two variants per region revealed only slight superiority of the local model (macro-F1: 92%) over the global model (macro-F1: 91%) in Indonesia with no difference in overall accuracy (OA) at 94%. In Central Africa, the results of the global and local model are the same in both metrics (OA: 84%, macro-F1: 82%). Overall, the outcome demonstrates the global model’s ability to generalize despite high dissimilarities between the research areas.

scholarly journals Classification of Very-High-Spatial-Resolution Aerial Images Based on Multiscale Features with Limited Semantic Information

2021 ◽  
Vol 13 (3) ◽  
pp. 364
Author(s):  
Han Gao ◽  
Jinhui Guo ◽  
Peng Guo ◽  
Xiuwan Chen
Keyword(s):  
Deep Learning ◽  
Land Cover ◽  
Spatial Resolution ◽  
Large Scale ◽  
High Spatial Resolution ◽  
Training Data ◽  
Aerial Images ◽  
Rural Landscapes ◽  
Feature Representations ◽  
Object Based

Recently, deep learning has become the most innovative trend for a variety of high-spatial-resolution remote sensing imaging applications. However, large-scale land cover classification via traditional convolutional neural networks (CNNs) with sliding windows is computationally expensive and produces coarse results. Additionally, although such supervised learning approaches have performed well, collecting and annotating datasets for every task are extremely laborious, especially for those fully supervised cases where the pixel-level ground-truth labels are dense. In this work, we propose a new object-oriented deep learning framework that leverages residual networks with different depths to learn adjacent feature representations by embedding a multibranch architecture in the deep learning pipeline. The idea is to exploit limited training data at different neighboring scales to make a tradeoff between weak semantics and strong feature representations for operational land cover mapping tasks. We draw from established geographic object-based image analysis (GEOBIA) as an auxiliary module to reduce the computational burden of spatial reasoning and optimize the classification boundaries. We evaluated the proposed approach on two subdecimeter-resolution datasets involving both urban and rural landscapes. It presented better classification accuracy (88.9%) compared to traditional object-based deep learning methods and achieves an excellent inference time (11.3 s/ha).

scholarly journals Improving 3-m Resolution Land Cover Mapping through Efficient Learning from an Imperfect 10-m Resolution Map

2020 ◽  
Vol 12 (9) ◽  
pp. 1418
Author(s):  
Runmin Dong ◽  
Cong Li ◽  
Haohuan Fu ◽  
Jie Wang ◽  
Weijia Li ◽  
...  
Keyword(s):  
Land Cover ◽  
Training Data ◽  
Training Dataset ◽  
Land Cover Mapping ◽  
Remotely Sensed Data ◽  
Large Area ◽  
National Scale ◽  
Substantial Progress ◽  
Efficient Learning ◽  
Land Cover Maps

Substantial progress has been made in the field of large-area land cover mapping as the spatial resolution of remotely sensed data increases. However, a significant amount of human power is still required to label images for training and testing purposes, especially in high-resolution (e.g., 3-m) land cover mapping. In this research, we propose a solution that can produce 3-m resolution land cover maps on a national scale without human efforts being involved. First, using the public 10-m resolution land cover maps as an imperfect training dataset, we propose a deep learning based approach that can effectively transfer the existing knowledge. Then, we improve the efficiency of our method through a network pruning process for national-scale land cover mapping. Our proposed method can take the state-of-the-art 10-m resolution land cover maps (with an accuracy of 81.24% for China) as the training data, enable a transferred learning process that can produce 3-m resolution land cover maps, and further improve the overall accuracy (OA) to 86.34% for China. We present detailed results obtained over three mega cities in China, to demonstrate the effectiveness of our proposed approach for 3-m resolution large-area land cover mapping.

scholarly journals Examining Land Cover and Greenness Dynamics in Hangzhou Bay in 1985–2016 Using Landsat Time-Series Data

2017 ◽  
Vol 10 (2) ◽  
pp. 32 ◽  
Author(s):  
Dengqiu Li ◽  
Dengsheng Lu ◽  
Ming Wu ◽  
Xuexin Shao ◽  
Jinhong Wei
Keyword(s):  
Time Series ◽  
Land Cover ◽  
Time Series Data ◽  
Series Data ◽  
Hangzhou Bay

scholarly journals Deep learning of stochastic contagion dynamics on complex networks

2020 ◽  
Author(s):  
Charles Murphy ◽  
Edward Laurence ◽  
Antoine Allard
Keyword(s):  
Deep Learning ◽  
Network Architecture ◽  
Time Series Data ◽  
Training Data ◽  
Series Data ◽  
Neural Network Architecture ◽  
Complementary Approach ◽  
Quantitative Accuracy ◽  
Simplifying Assumptions ◽  
Effective Models

Abstract Forecasting the evolution of contagion dynamics is still an open problem to which mechanistic models only offer a partial answer. To remain mathematically and/or computationally tractable, these models must rely on simplifying assumptions, thereby limiting the quantitative accuracy of their predictions and the complexity of the dynamics they can model. Here, we propose a complementary approach based on deep learning where the effective local mechanisms governing a dynamic are learned automatically from time series data. Our graph neural network architecture makes very few assumptions about the dynamics, and we demonstrate its accuracy using stochastic contagion dynamics of increasing complexity on static and temporal networks. By allowing simulations on arbitrary network structures, our approach makes it possible to explore the properties of the learned dynamics beyond the training data. Our results demonstrate how deep learning offers a new and complementary perspective to build effective models of contagion dynamics on networks.

scholarly journals USING SMOS PASSIVE MICROWAVE DATA TO DEVELOP SMAP FREEZE/THAW ALGORITHMS ADAPTED FOR THE CANADIAN SUBARCTIC

2015 ◽  
Vol XL-1-W5 ◽  
pp. 365-368 ◽  
Author(s):  
P. Kalantari ◽  
M. Bernier ◽  
K. C. McDonal ◽  
J. Poulin
Keyword(s):  
Land Cover ◽  
Time Series Data ◽  
Passive Microwave ◽  
Soil Freezing ◽  
Series Data ◽  
Freezing And Thawing ◽  
Freeze Thaw ◽  
Thaw Cycle ◽  
Freeze Thaw Cycle ◽  
The Impact

Seasonal terrestrial Freeze/Thaw cycle in Northern Quebec Tundra (Nunavik) was determined and evaluated with passive microwave observations. SMOS time series data were analyzed to examine seasonal variations of soil freezing, and to assess the impact of land cover on the Freeze/Thaw cycle. Furthermore, the soil freezing maps derived from SMOS observations were compared to field survey data in the region near Umiujaq. The objective is to develop algorithms to follow the seasonal cycle of freezing and thawing of the soil adapted to Canadian subarctic, a territory with a high complexity of land cover (vegetation, soil, and water bodies). Field data shows that soil freezing and thawing dates vary much spatially at the local scale in the Boreal Forest and the Tundra. The results showed a satisfactory pixel by pixel mapping for the daily soil state monitoring with a > 80% success rate with in situ data for the HH and VV polarizations, and for different land cover. The average accuracies are 80% and 84% for the soil freeze period, and soil thaw period respectively. The comparison is limited because of the small number of validation pixels.

scholarly journals HYBRID ACQUISITION OF HIGH QUALITY TRAINING DATA FOR SEMANTIC SEGMENTATION OF 3D POINT CLOUDS USING CROWD-BASED ACTIVE LEARNING

2020 ◽  
Vol V-2-2020 ◽  
pp. 501-508
Author(s):  
M. Kölle ◽  
V. Walter ◽  
S. Schmohl ◽  
U. Soergel
Keyword(s):  
Active Learning ◽  
Semantic Segmentation ◽  
Point Clouds ◽  
Human Interaction ◽  
Training Data ◽  
Semantic Interpretation ◽  
Training Dataset ◽  
3D Point Clouds ◽  
Semantic Labeling ◽  
3D Cnn

Abstract. Automated semantic interpretation of 3D point clouds is crucial for many tasks in the domain of geospatial data analysis. For this purpose, labeled training data is required, which has often to be provided manually by experts. One approach to minimize effort in terms of costs of human interaction is Active Learning (AL). The aim is to process only the subset of an unlabeled dataset that is particularly helpful with respect to class separation. Here a machine identifies informative instances which are then labeled by humans, thereby increasing the performance of the machine. In order to completely avoid involvement of an expert, this time-consuming annotation can be resolved via crowdsourcing. Therefore, we propose an approach combining AL with paid crowdsourcing. Although incorporating human interaction, our method can run fully automatically, so that only an unlabeled dataset and a fixed financial budget for the payment of the crowdworkers need to be provided. We conduct multiple iteration steps of the AL process on the ISPRS Vaihingen 3D Semantic Labeling benchmark dataset (V3D) and especially evaluate the performance of the crowd when labeling 3D points. We prove our concept by using labels derived from our crowd-based AL method for classifying the test dataset. The analysis outlines that by labeling only 0:4% of the training dataset by the crowd and spending less than 145 $, both our trained Random Forest and sparse 3D CNN classifier differ in Overall Accuracy by less than 3 percentage points compared to the same classifiers trained on the complete V3D training set.

Sign in / Sign up

Export Citation Format

Share Document