Identification of Crop Type in Crowdsourced Road View Photos with Deep Convolutional Neural Network

In situ ground truth data are an important requirement for producing accurate cropland type map, and this is precisely what is lacking at vast scales. Although volunteered geographic information (VGI) has been proven as a possible solution for in situ data acquisition, processing and extracting valuable information from millions of pictures remains challenging. This paper targets the detection of specific crop types from crowdsourced road view photos. A first large, public, multiclass road view crop photo dataset named iCrop was established for the development of crop type detection with deep learning. Five state-of-the-art deep convolutional neural networks including InceptionV4, DenseNet121, ResNet50, MobileNetV2, and ShuffleNetV2 were employed to compare the baseline performance. ResNet50 outperformed the others according to the overall accuracy (87.9%), and ShuffleNetV2 outperformed the others according to the efficiency (13 FPS). The decision fusion schemes major voting was used to further improve crop identification accuracy. The results clearly demonstrate the superior accuracy of the proposed decision fusion over the other non-fusion-based methods in crop type detection of imbalanced road view photos dataset. The voting method achieved higher mean accuracy (90.6–91.1%) and can be leveraged to classify crop type in crowdsourced road view photos.

Download Full-text

In-Season Major Crop-Type Identification for US Cropland from Landsat Images Using Crop-Rotation Pattern and Progressive Data Classification

Agriculture ◽

10.3390/agriculture9010017 ◽

2019 ◽

Vol 9 (1) ◽

pp. 17 ◽

Cited By ~ 5

Author(s):

Md. Shahinoor Rahman ◽

Liping Di ◽

Eugene Yu ◽

Chen Zhang ◽

Hossain Mohiuddin

Keyword(s):

Image Classification ◽

Crop Rotation ◽

Ground Truth ◽

The United States ◽

Classification Models ◽

Landsat Images ◽

Ground Truth Data ◽

Rotation Pattern ◽

Crop Type ◽

Crop Types

Crop type information at the field level is vital for many types of research and applications. The United States Department of Agriculture (USDA) provides information on crop types for US cropland as a Cropland Data Layer (CDL). However, CDL is only available at the end of the year after the crop growing season. Therefore, CDL is unable to support in-season research and decision-making regarding crop loss estimation, yield estimation, and grain pricing. The USDA mostly relies on field survey and farmers’ reports for the ground truth to train image classification models, which is one of the major reasons for the delayed release of CDL. This research aims to use trusted pixels as ground truth to train classification models. Trusted pixels are pixels which follow a specific crop rotation pattern. These trusted pixels are used to train image classification models for the classification of in-season Landsat images to identify major crop types. Six different classification algorithms are investigated and tested to select the best algorithm for this study. The Random Forest algorithm stands out among selected algorithms. This study classified Landsat scenes between May and mid-August for Iowa. The overall agreements of classification results with CDL in 2017 are 84%, 94%, and 96% for May, June, and July, respectively. The classification accuracies have been assessed through 683 ground truth data points collected from the fields. The overall accuracies of single date multi-band image classification are 84%, 89% and 92% for May, June, and July, respectively. The result also shows higher accuracy (94–95%) can be achieved through multi-date image classification compared to single date image classification.

Download Full-text

Consistency of Radiometric Satellite Data over Lakes and Coastal Waters with Local Field Measurements

Remote Sensing ◽

10.3390/rs12040616 ◽

2020 ◽

Vol 12 (4) ◽

pp. 616 ◽

Cited By ~ 3

Author(s):

Krista Alikas ◽

Ilmar Ansko ◽

Viktor Vabson ◽

Ave Ansper ◽

Kersti Kangro ◽

...

Keyword(s):

Coastal Waters ◽

Atmospheric Correction ◽

Field Measurements ◽

Ground Truth ◽

Local Data ◽

Ground Truth Data ◽

In Situ Data ◽

Global Coverage

The Sentinel-3 mission launched its first satellite Sentinel-3A in 2016 to be followed by Sentinel-3B and Sentinel-3C to provide long-term operational measurements over Earth. Sentinel-3A and 3B are in full operational status, allowing global coverage in less than two days, usable to monitor optical water quality and provide data for environmental studies. However, due to limited ground truth data, the product quality has not yet been analyzed in detail with the fiducial reference measurement (FRM) dataset. Here, we use the fully characterized ground truth FRM dataset for validating Sentinel-3A Ocean and Land Colour Instrument (OLCI) radiometric products over optically complex Estonian inland waters and Baltic Sea coastal areas. As consistency between satellite and local data depends on uncertainty in field measurements, filtering of the in situ data has been made based on the uncertainty for the final comparison. We have compared various atmospheric correction methods and found POLYMER (POLYnomial-based algorithm applied to MERIS) to be most suitable for optically complex waters under study in terms of product accuracy, amount of usable data and also being least influenced by the adjacency effect.

Download Full-text

Assessment with Controlled In-Situ Data of the Dependence of L-Band Radiometry on Sea-Ice Thickness

Remote Sensing ◽

10.3390/rs12040650 ◽

2020 ◽

Vol 12 (4) ◽

pp. 650

Author(s):

Pablo Sánchez-Gámez ◽

Carolina Gabarro ◽

Antonio Turiel ◽

Marcos Portabella

Keyword(s):

Soil Moisture ◽

Sea Ice ◽

Brightness Temperature ◽

European Space Agency ◽

Ground Truth ◽

Ice Thickness ◽

Sea Ice Thickness ◽

In Situ Data ◽

L Band

The European Space Agency (ESA) Soil Moisture and Ocean Salinity (SMOS) and the National Aeronautics and Space Administration (NASA) Soil Moisture Active Passive (SMAP) missions are providing brightness temperature measurements at 1.4 GHz (L-band) for about 10 and 4 years respectively. One of the new areas of geophysical exploitation of L-band radiometry is on thin (i.e., less than 1 m) Sea Ice Thickness (SIT), for which theoretical and empirical retrieval methods have been proposed. However, a comprehensive validation of SIT products has been hindered by the lack of suitable ground truth. The in-situ SIT datasets most commonly used for validation are affected by one important limitation: They are available mainly during late winter and spring months, when sea ice is fully developed and the thickness probability density function is wider than for autumn ice and less representative at the satellite spatial resolution. Using Upward Looking Sonar (ULS) data from the Woods Hole Oceanographic Institution (WHOI), acquired all year round, permits overcoming the mentioned limitation, thus improving the characterization of the L-band brightness temperature response to changes in thin SIT. State-of-the-art satellite SIT products and the Cumulative Freezing Degree Days (CFDD) model are verified against the ULS ground truth. The results show that the L-band SIT can be meaningfully retrieved up to 0.6 m, although the signal starts to saturate at 0.3 m. In contrast, despite the simplicity of the CFDD model, its predicted SIT values correlate very well with the ULS in-situ data during the sea ice growth season. The comparison between the CFDD SIT and the current L-band SIT products shows that both the sea ice concentration and the season are fundamental factors influencing the quality of the thickness retrieval from L-band satellites.

Download Full-text

A Sentinel-1 based European crop parcel map using 2018 in-situ LUCAS Copernicus observations

10.5194/egusphere-egu2020-18703 ◽

2020 ◽

Author(s):

Verhegghen Astrid ◽

d'Andrimont Raphaël ◽

Lemoine Guido ◽

Strobl Peter ◽

van der Velde Marijn

Keyword(s):

Remote Sensing ◽

European Union ◽

Land Cover ◽

Growing Season ◽

Earth Observation ◽

Soft Wheat ◽

High Quality ◽

In Situ Data ◽

Crop Type

Efficient near-real time and wall-to-wall land monitoring is now possible with unprecedented detail because of the fleet of Copernicus Sentinel satellites. This remote sensing paradigm is the consequence of the freely accessible, global, Copernicus data, combined with affordable cloud computing. However, to translate this capacity in accurate products, and to truly benefit from the high spatial detail (~10m) and temporal resolution (~5 days in constellation) of the Sentinels 1 and 2, high quality and timely in-situ data remains crucial. Robust operational monitoring systems are in need of both training and validation data.&#160;Here, we demonstrate the potential of Sentinel 1 observations and complementary high-quality in-situ data to generate a crop type map at continental scale. In 2018, the Land Cover and Land Use Area frame Survey (LUCAS) carried out in the European Union contained a specific Copernicus module corresponding to 93.091 polygons surveyed in-situ. In contrast to the usual LUCAS point observation, the Copernicus protocol provides data on the extent of homogeneous land cover for a maximum size of 100 x 100 m, making it meaningful for remote sensing applications. After filtering the polygons to retrieve only high quality sample, a sample was selected to explore the accuracy of crop type maps at different moments of the 2018 growing season over Europe. The time series of 10 days VV and VH were classified using Random Forest models. The crops that were mapped correspond to the 13 major crops in Europe and are those that are monitored and forecast by the JRC MARS activities (soft wheat, maize, rapeseed, barley, potatoes, ...). Overall, reasonable accuracies were obtained (~80%). Although no a priori parcel delineation was used, it was encouraging to observe the relative homogeneity of pixel classification results within the same parcel. In the context of forecasting, we specifically assessed at what time in the growing season accuracies moved beyond a set threshold for the different crops. This ranged from May for winter crops such as soft wheat, and September for summer crops such as maize.&#160;Our results contribute to the discussion regarding the usefulness, benefits, as well as weaknesses, of the newly acquired LUCAS Copernicus data. Doing so, this study demonstrates the potential of in-situ surveys such as LUCAS Copernicus module&#160; specifically targeted for Earth Observation applications. Future improvements to the LUCAS Copernicus survey methodology are suggested. Importantly, now that LUCAS has been postponed to 2022, and aligned with the Copernicus space program, we advocate for a European Union wide systematic and representative in-situ sample campaign relevant for Earth Observation applications, beyond the traditional LUCAS survey.&#160;

Download Full-text

Harmonized in situ JECAM datasets for agricultural land use mapping and monitoring in tropical countries

10.5194/essd-2021-125 ◽

2021 ◽

Author(s):

Audrey Jolivot ◽

Valentine Lebourgeois ◽

Mael Ameline ◽

Valérie Andriamanga ◽

Beatriz Bellón ◽

...

Keyword(s):

Land Use ◽

Agricultural Land ◽

Satellite Image ◽

Farming Systems ◽

Agricultural Land Use ◽

In Situ Data ◽

Agricultural Census ◽

Crop Types ◽

The Tropics

Abstract. The availability of crop type reference datasets for satellite image classification is very limited for complex agricultural systems as observed in developing and emerging countries. Indeed, agricultural land use is very dynamic, agricultural census are often poorly georeferenced, and crop types are difficult to photo-interpret directly from satellite imagery. In this paper, we present nine datasets collected in a standardized manner between 2013 and 2020 in seven tropical and subtropical countries within the framework of the international JECAM (Joint Experiment for Crop Assessment and Monitoring) initiative. These quality-controlled datasets are distinguished by in situ data collected at field scale by local experts, with precise geographic coordinates, and following a common protocol. Altogether, the datasets completed 27 074 polygons (20 257 crop and 6 817 non-crop) documented by detailed keywords. These datasets can be used to produce and validate agricultural land use maps in the tropics, but also, to assess the performances and the robustness of classification methods of cropland and crop types/practices in a large range of tropical farming systems. The dataset is available at https://doi.org/10.18167/DVN1/P7OLAP.

Download Full-text

The Dangers of Post-hoc Interpretability: Unjustified Counterfactual Explanations

Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence ◽

10.24963/ijcai.2019/388 ◽

2019 ◽

Cited By ~ 4

Author(s):

Thibault Laugel ◽

Marie-Jeanne Lesot ◽

Christophe Marsala ◽

Xavier Renard ◽

Marcin Detyniecki

Keyword(s):

State Of The Art ◽

Ground Truth ◽

Black Box ◽

Box Model ◽

Ground Truth Data ◽

Actual Knowledge ◽

Post Hoc

Post-hoc interpretability approaches have been proven to be powerful tools to generate explanations for the predictions made by a trained black-box model. However, they create the risk of having explanations that are a result of some artifacts learned by the model instead of actual knowledge from the data. This paper focuses on the case of counterfactual explanations and asks whether the generated instances can be justified, i.e. continuously connected to some ground-truth data. We evaluate the risk of generating unjustified counterfactual examples by investigating the local neighborhoods of instances whose predictions are to be explained and show that this risk is quite high for several datasets. Furthermore, we show that most state of the art approaches do not differentiate justified from unjustified counterfactual examples, leading to less useful explanations.

Download Full-text

A tool for semi-automatic ground truth annotation of traffic videos

Electronic Imaging ◽

10.2352/issn.2470-1173.2020.16.avm-150 ◽

2020 ◽

Vol 2020 (16) ◽

pp. 200-1-200-7

Author(s):

Florian Groh ◽

Dominik Schörkhuber ◽

Margrit Gelautz

Keyword(s):

User Study ◽

State Of The Art ◽

User Interaction ◽

Ground Truth ◽

Automatic Annotation ◽

Video Traffic ◽

Ground Truth Data ◽

Lighting Conditions ◽

Ground Truth Generation ◽

Dynamic Lighting

We have developed a semi-automatic annotation tool – “CVL Annotator” – for bounding box ground truth generation in videos. Our research is particularly motivated by the need for reference annotations of challenging nighttime traffic scenes with highly dynamic lighting conditions due to reflections, headlights and halos from oncoming traffic. Our tool incorporates a suite of different state-of-the-art tracking algorithms in order to minimize the amount of human input necessary to generate high-quality ground truth data. We focus our user interface on the premise of minimizing user interaction and visualizing all information relevant to the user at a glance. We perform a preliminary user study to measure the amount of time and clicks necessary to produce ground truth annotations of video traffic scenes and evaluate the accuracy of the final annotation results.

Download Full-text

Targeted Grassland Monitoring at Parcel Level Using Sentinels, Street-Level Images and Field Observations

Remote Sensing ◽

10.3390/rs10081300 ◽

2018 ◽

Vol 10 (8) ◽

pp. 1300 ◽

Cited By ~ 13

Author(s):

Raphaël d’Andrimont ◽

Guido Lemoine ◽

Marijn van der Velde

Keyword(s):

High Resolution ◽

Collection Efficiency ◽

Ground Truth ◽

Bare Soil ◽

Agricultural Policies ◽

High Definition ◽

Street Level ◽

In Situ Data ◽

Crop Type ◽

Efficiency And Effectiveness

The introduction of high-resolution Sentinels combined with the use of high-quality digital agricultural parcel registration systems is driving the move towards at-parcel agricultural monitoring. The European Union’s Common Agricultural Policy (CAP) has introduced the concept of CAP monitoring to help simplify the management and control of farmers’ parcel declarations for area support measures. This study proposes a proof of concept of this monitoring approach introducing and applying the concept of ‘markers’. Using Sentinel-1- and -2-derived (S1 and S2) markers, we evaluate parcels declared as grassland in the Gelderse Vallei in the Netherlands covering more than 15,000 parcels. The satellite markers—respectively based on crop-type deep learning classification using S1 backscattering and coherence data and on detecting bare soil with S2 during the growing season—aim to identify grassland-declared parcels for which (1) the marker suggests another crop type or (2) which appear to have been ploughed during the year. Subsequently, a field-survey was carried out in October 2017 to target the parcels identified and to build a relevant ground-truth sample of the area. For the latter purpose, we used a high-definition camera mounted on the roof of a car to continuously sample geo-tagged digital imagery, as well as an app-based approach to identify the targeted fields. Depending on which satellite-based marker or combination of markers is used, the number of parcels identified ranged from 2.57% (marked by both the S1 and S2 markers) to 17.12% of the total of 11,773 parcels declared as grassland. After confirming with the ground-truth, parcels flagged by the combined S1 and S2 marker were robustly detected as non-grassland parcels (F-score = 0.9). In addition, the study demonstrated that street-level imagery collection could improve collection efficiency by a factor seven compared to field visits (1411 parcels/day vs. 217 parcels/day) while keeping an overall accuracy of about 90% compared to the ground-truth. This proposed way of collecting in situ data is suitable for the training and validating of high resolution remote sensing approaches for agricultural monitoring. Timely country-wide wall-to-wall parcel-level monitoring and targeted in-season parcel surveying will increase the efficiency and effectiveness of monitoring and implementing agricultural policies.

Download Full-text

Combining MODIS, AVHRR and in situ data for evapotranspiration estimation over heterogeneous landscape of the Tibetan Plateau

Atmospheric Chemistry and Physics ◽

10.5194/acp-14-1507-2014 ◽

2014 ◽

Vol 14 (3) ◽

pp. 1507-1515 ◽

Cited By ~ 28

Author(s):

Y. Ma ◽

Z. Zhu ◽

L. Zhong ◽

B. Wang ◽

C. Han ◽

...

Keyword(s):

Tibetan Plateau ◽

Ground Truth ◽

The Tibetan Plateau ◽

Evaporative Fraction ◽

Modis Data ◽

In Situ Data ◽

Heterogeneous Landscape ◽

Wide Range ◽

Avhrr Data

Abstract. In this study, a parameterization method based on MODIS (Moderate Resolution Imaging Spectroradiometer) data, AVHRR (Advanced Very High-Resolution Radiometer) data and in situ data is introduced and tested for estimating the regional evaporative fraction Λ over a heterogeneous landscape. As a case study, the algorithm was applied to the Tibetan Plateau (TP) area. Eight MODIS data images (17 January, 14 April, 23 July and 16 October in 2003; 30 January, 15 April, 1 August and 25 October in 2007) and four AVHRR data images (17 January, 14 April, 23 July and 16 October in 2003) were used in this study to compare winter, spring, summer and autumn values and for annual variation analysis. The results were validated using the "ground truth" measured at Tibetan Observation and Research Platform (TORP) and the CAMP/Tibet (CEOP (Coordinated Enhanced Observing Period) Asia-Australia Monsoon Project (CAMP) on the Tibetan Plateau) meteorological stations. The results show that the estimated evaporative fraction Λ in the four different seasons over the TP is in clear accordance with the land surface status. The Λ fractions show a wide range due to the strongly contrasting surface features found on the TP. Also, the estimated Λ values are in good agreement with "ground truth" measurements, and their absolute percentage difference (APD) is less than 10.0% at the validation sites. The AVHRR data were also in agreement with the MODIS data, with the latter usually displaying a higher level of accuracy. It was therefore concluded that the proposed algorithm was successful in retrieving the evaporative fraction Λ using MODIS, AVHRR and in situ data over the TP. MODIS data are the most accurate and should be used widely in evapotranspiration (ET) research in this region.

Download Full-text

Synergy between optical imaging radiometry and radar altimetry for inland waters: an experience with Sentinel-3 on the Nasser Lake

10.5194/egusphere-egu2020-18804 ◽

2020 ◽

Author(s):

Andrea Scozzari ◽

Stefano Vignudelli ◽

Mohamed Elsahabi ◽

Neama Galal ◽

Marwa Khairy ◽

...

Keyword(s):

Land Surface ◽

Morphological Changes ◽

Morphological Characteristics ◽

Ground Truth ◽

Water Levels ◽

Radar Altimeter ◽

Radar Altimetry ◽

Ground Truth Data ◽

In Situ Data ◽

Nasser Lake

It is currently well known that a combination of stressors, such as climate change, human activities and new infrastructures might influence the storage capacity of strategic surface water reservoirs at a global level.The Nasser Lake is the biggest and most important lake in Egypt, located in the southern part of the Nile River in Upper Egypt. The expected impact of the Grand Ethiopian Renaissance Dam (GERD) on the future availability of the Nile water, together with the significant and rapid water level variations and sedimentation processes, make the Nasser Lake a particularly challenging place to be monitored in the next years.This work describes a preliminary study on the possible usage of the imaging radiometer SLSTR (Sea and Land Surface Temperature Radiometer) onboard Sentinel-3 for estimating water coverage extent in inland water contexts, in synergy with radar altimetry measurements provided by the SRAL (Synthetic aperture Radar ALtimeter) instrument. In particular, this work wants to exploit the simultaneous acquisition offered by SRAL and SLSTR instruments hosted by the Sentinel-3A/B platform.We introduce an alternative technique to the classical calculation of the whole water extent based on high-resolution imagery, essentially intended for the application to wide-swath short-revisit sensors. The proposed approach starts from the hypothesis that a much-reduced subset of pixels may carry enough information for assessing the status of the observed water body by estimating the water coverage percent within each single pixel. Such an assumption can rely only on the radiometric performance of the instrument, SLSTR in this case.The timeseries of water levels by the SRAL instrument were obtained by using the 20 Hz product generated by the SARvatore processor run on the ESA GPOD (Grid Processing On Demand) platform. A timeseries derived from SLSTR measurements has been generated by a simple feature extraction technique, based on the selection of pixels exhibiting the highest variability of the collected radiance. As expected, this subset essentially identifies particular spots on the coastlines of the target, as a consequence of its morphological characteristics.Preliminary results show a promising relationship between the timeseries generated by the two independent measurements and between the available in situ data as well. Under the hypothesis of a time-invariant system (i.e., characterised by no significant morphological changes), once an area-level-volume relationship is identified, volume estimations can be inferred by either altimetric or radiometric measurements per se.Thus, the simultaneous observation by the two instruments represents a relevant opportunity for cross-validating the acquired data. Moreover, the approximation experimented in this work gives the perspective of a very light computational process for expedite water storage estimations in large surface reservoirs, provided that the natural system is fully identified on the basis of ground-truth data.

Download Full-text