Recently proposed methods for weakly-supervised semantic segmentation have achieved impressive performance in predicting pixel classes despite being trained with only image labels which lack ...positional information. Because image annotations are cheaper and quicker to generate, weak supervision is more practical than full supervision for training segmentation algorithms. These methods have been predominantly developed to solve the background separation and partial segmentation problems presented by natural scene images and it is unclear whether they can be simply transferred to other domains with different characteristics, such as histopathology and satellite images, and still perform well. This paper evaluates state-of-the-art weakly-supervised semantic segmentation methods on natural scene, histopathology, and satellite image datasets and analyzes how to determine which method is most suitable for a given dataset. Our experiments indicate that histopathology and satellite images present a different set of problems for weakly-supervised semantic segmentation than natural scene images, such as ambiguous boundaries and class co-occurrence. Methods perform well for datasets they were developed on, but tend to perform poorly on other datasets. We present some practical techniques for these methods on unseen datasets and argue that more work is needed for a generalizable approach to weakly-supervised semantic segmentation. Our full code implementation is available on GitHub:
https://github.com/lyndonchan/wsss-analysis
.
The coordination of humanitarian relief, e.g. in a natural disaster or a conflict situation, is often complicated by a scarcity of data to inform planning. Remote sensing imagery, from satellites or ...drones, can give important insights into conditions on the ground, including in areas which are difficult to access. Applications include situation awareness after natural disasters, structural damage assessment in conflict, monitoring human rights violations or population estimation in settlements. We review machine learning approaches for automating these problems, and discuss their potential and limitations. We also provide a case study of experiments using deep learning methods to count the numbers of structures in multiple refugee settlements in Africa and the Middle East. We find that while high levels of accuracy are possible, there is considerable variation in the characteristics of imagery collected from different sensors and regions. In this, as in the other applications discussed in the paper, critical inferences must be made from a relatively small amount of pixel data. We, therefore, consider that using machine learning systems as an augmentation of human analysts is a reasonable strategy to transition from current fully manual operational pipelines to ones which are both more efficient and have the necessary levels of quality control.
This article is part of a discussion meeting issue 'The growing ubiquity of algorithms in society: implications, impacts and innovations'.
The Simple Algorithm for Yield estimates (SAFY) is a crop yield model that simulates crop growth and biomass accumulation at a daily time step. Parameters in the SAFY model can be determined from ...literature, in situ measurements, or optical remote sensing data through data assimilation. For effective determination of parameters, optical remote sensing data need to be acquired at high spatial and high temporal resolutions. However, this is challenging due to interference of cloud cover and rather long revisiting cycles of high resolution satellite sensors. Spatio-temporal fusion of multi-source remote sensing data may represent a feasible solution. Here, crop phenology-related parameters in the SAFY model were derived using an improved Two-Step Filtering (TSF) model from remote sensing data generated through spatio-temporal fusion of Landsat-8 and Moderate Resolution Imaging Spectroradiometer (MODIS) data. Remaining parameters were determined through an optimization procedure using the same dataset. The SAFY model was then used for dry aboveground biomass and yield estimation at a subfield scale for corn (Zea mays) and soybean (Glycine max). The results show that the improved TSF method is able to determine crop phenology stages with an error of <5 days. After calibration, the SAFY model can reproduce daily Green Leaf Area Index (GLAI) effectively throughout the growing season and estimate crop biomass and yield accurately at a subfield scale using three Landsat-8 and 10 MODIS images acquired for the season. This approach improves the accuracy of biomass estimation by about 4% in relative Root Mean Square Error (RRMSE), compared with the SAFY model without forcing the phenology-related parameters. The RMSE of yield estimation is 146.33 g/m2 for corn and 82.86 g/m2 for soybean. The proposed framework is applicable for local-scale or field-scale phenology detection and yield estimation.
Display omitted
•Improves the Two-Step Filtering method for phenology detection•Calibrates the Simple Algorithm for Yield estimates model for corn and soybean•Estimates the biomass and yield accurately at a subfield scale•A good correlation is found between effective light use efficiency and fAPARmax.
Global trends in satellite-based emergency mapping Voigt, Stefan; Giulio-Tonolo, Fabio; Lyons, Josh ...
Science (American Association for the Advancement of Science),
07/2016, Letnik:
353, Številka:
6296
Journal Article
Recenzirano
Over the past 15 years, scientists and disaster responders have increasingly used satellite-based Earth observations for global rapid assessment of disaster situations. We review global trends in ...satellite rapid response and emergency mapping from 2000 to 2014, analyzing more than 1000 incidents in which satellite monitoring was used for assessing major disaster situations. We provide a synthesis of spatial patterns and temporal trends in global satellite emergency mapping efforts and show that satellite-based emergency mapping is most intensively deployed in Asia and Europe and follows well the geographic, physical, and temporal distributions of global natural disasters. We present an outlook on the future use of Earth observation technology for disaster response and mitigation by putting past and current developments into context and perspective.
Rapid and extensive urbanization has adversely impacted humans and ecological entities in the recent decades through a decrease in surface permeability and the emergence of Urban Heat Islands (UHI). ...While detailed and continuous assessments of surface permeability and UHI are crucial for urban planning and management of landuse zones, they mostly involve time consuming and expensive field studies and single sensor derived large scale aerial and satellite imageries. We demonstrated the advantage of fusing imageries from multiple sensors for landuse and landcover (LULC) change assessments as well as for assessing surface permeability and temperature and UHI emergence in a fast growing city, i.e. Tirunelveli, Tamilnadu, India. IRS-LISSIII and Landsat-7 ETM+ imageries were fused for 2007 and 2017, and classified using a Rotation Forest (RF) algorithm. Surface permeability and temperature were then quantified using Soil-Adjusted Vegetation Index (SAVI) and Land Surface Temperature (LST) index, respectively. Finally, we assessed the relationship between SAVI and LST for entire Tirunelveli as well as for each LULC zone, and also detected UHI emergence hot spots using a SAVI-LST combined metric. Our fused images exhibited higher classification accuracies, i.e. overall kappa coefficient values, than non-fused images. We observed an overall increase in the coverage of urban (dry, real estate plots and built-up) areas, while a decrease for vegetated (cropland and forest) areas in Tirunelveli between 2007 and 2017. The SAVI values indicated an extensive decrease in surface permeability for Tirunelveli overall and also for almost all LULC zones. The LST values showed an overall increase of surface temperature in Tirunelveli with the highest increase for urban built-up areas between 2007 and 2017. LST also exhibited a strong negative association with SAVI. Southeastern built-up areas in Tirunelveli were depicted as a potential UHI hotspot, with a caution for the Western riparian zone for UHI emergence in 2017. Our results provide important metrics for surface permeability, temperature and UHI monitoring, and inform urban and zonal planning authorities about the advantages of satellite image fusion.
Photosynthesis of the Amazon rainforest plays an important role in the regional and global carbon cycles, but, despite considerable in situ and space-based observations, it has been intensely debated ...whether there is a dry-season increase in greenness and photosynthesis of the moist tropical Amazonian forests. Solar-induced chlorophyll fluorescence (SIF), which is emitted by chlorophyll, has a strong positive linear relationship with photosynthesis at the canopy scale. Recent advancements have allowed us to observe SIF globally with Earth observation satellites. Here we show that forest SIF did not decrease in the early dry season and increased substantially in the late dry season and early part of wet season, using SIF data from the Tropospheric Monitoring Instrument (TROPOMI), which has unprecedented spatial resolution and near-daily global coverage. Using in situ CO₂ eddy flux data, we also show that cloud cover rarely affects photosynthesis at TROPOMI’s midday overpass, a time when the forest canopy is most often light-saturated. The observed dry-season increases of forest SIF are not strongly affected by sun-sensor geometry, which was attributed as creating a pseudo dry-season green-up in the surface reflectance data. Our results provide strong evidence that greenness, SIF, and photosynthesis of the tropical Amazonian forest increase during the dry season.
Urban growth and decline occur every year and show changes in urban areas. Although various approaches to detect urban changes have been developed, they mainly use large-scale satellite imagery and ...socioeconomic factors in urban areas, which provides an overview of urban changes. However, since people explore places and notice changes daily at the street level, it would be useful to develop a method to identify urban changes at the street level and demonstrate whether urban growth or decline occurs there. Thus, this study seeks to use street-level panoramic images from Google Street View to identify urban changes and to develop a new way to evaluate the growth and decline of an urban area. After collecting Google Street View images year by year, we trained and developed a deep-learning model of an object detection process using the open-source software TensorFlow. By scoring objects and changes detected on a street from year to year, a map of urban growth and decline was generated for Midtown in Detroit, Michigan, USA. By comparing socioeconomic changes and the situations of objects and changes in Midtown, the proposed method is shown to be helpful for analyzing urban growth and decline by using year-by-year street view images.
•Soybean yield at municipality-level was forecasted using satellite and weather data.•LSTM neural networks outperformed conventional machine learning algorithms in soybean yield prediction.•The model ...accuracy decreased as we anticipated earlier dates of the predictions.•Soybean yield can be forecasted with MAE of 0.42 Mg ha−1 ~70 days before harvesting.
Soybean yield predictions in Brazil are of great interest for market behavior, to drive governmental policies and to increase global food security. In Brazil soybean yield data generally demand various revisions through the following months after harvest suggesting that there is space for improving the accuracy and the time of yield predictions. This study presents a novel model to perform in-season (“near real-time”) soybean yield forecasts in southern Brazil using Long-Short Term Memory (LSTM), Neural Networks, satellite imagery and weather data. The objectives of this study were to: (i) compare the performance of three different algorithms (multivariate OLS linear regression, random forest and LSTM neural networks) for forecasting soybean yield using NDVI, EVI, land surface temperature and precipitation as independent variables, and (ii) evaluate how early (during the soybean growing season) this method is able to forecast yield with reasonable accuracy. Satellite and weather data were masked using a non-crop-specific layer with field boundaries obtained from the Rural Environment Registry that is mandatory for all farmers in Brazil. Main outcomes from this study were: (i) soybean yield forecasts at municipality-scale with a mean absolute error (MAE) of 0.24 Mg ha−1 at DOY 64 (march 5) (ii) a superior performance of the LSTM neural networks relative to the other algorithms for all the forecast dates except DOY 16 where multivariate OLS linear regression provided the best performance, and (iii) model performance (e.g., MAE) for yield forecast decreased when predictions were performed earlier in the season, with MAE increasing from 0.24 Mg ha−1 to 0.42 Mg ha−1 (last values from OLS regression) when forecast timing changed from DOY 64 (March 5) to DOY 16 (January 6). This research portrays the benefits of integrating statistical techniques, remote sensing, weather to field survey data in order to perform more reliable in-season soybean yield forecasts.
Abstract
In this work, a calibration method for deriving a sensitivity of field pyranometers using Himawari satellite image data was developed. The purposed method representing a relation between ...10–bit digital counts ranging between 0–1023 from satellite data and global solar irradiance of a standard pyranometer measured at Nakhon Pathom (13.82 °N, 100.04 °E) was investigated and it was found to be an exponential relation. Afterward, this model was applied to estimate the sensitivity of field pyranometers installed at three meteorological monitoring stations located in the main regions of Thailand namely, Chaing Mai in the North (18.78 °N, 98.98 °E), Ubon Ratchathani in the Northeast (15.52 °N, 104.87 °E) and Songkhla in the South (7.20 °N, 100.60 °E) for the validation. The results showed that the sensitivity of field pyranometers derived from satellite data and that obtained from the standard outdoor calibration method of ISO9847 was in a very good agreement with a discrepancy between both datasets of about 1%.