wph earth observation - europa...2 case study 1 (pl) agriculture - crop recognition, mapping and...
Post on 14-Aug-2020
1 Views
Preview:
TRANSCRIPT
WPH – Earth observation Case Study 1 (PL) Agriculture-Crop recognition, mapping and monitoring
05 – 06 December 2018, Vienna, Austria 1
Marek Morze
2
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 1 – data collection a) collection of time series (X 2017 – IX 2018) Sentinel-1A/B SAR data in interferometric wide swath mode for crop recognition (provides continuous imagery day, night and all weather, revisit time – 6 days). Sentinel-1 working modes: • SM (Stripmap Mode) – acquires data with an 80 km swath at 5 m by 5 m spatial resolution, • IW (Interferometric Wideswath Mode) – acquires data with a 250 km swath at 5 m by 20 m spatial resolution, • EW (Extra-wide Swath Mode) – acquires data over a 400 km swath at 20 m by 40 m spatial resolution, • WV (Wave Mode) – acquires data in 20 km by 20 km area, at 5 m by 5 m under angle 23° and 36.5°.
3
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 1 – data collection b) collection of Sentinel-2A/B VNIR/SWIR images for the same period for image segmentation suport and calculating vegetation indexes, (wide-swath - 290 km, high-resolution, revisit time– 6 days), Sentinel-2 spectral bands: • four bands at 10 metres spatial resolution– 490 nm (B2), 560 nm (B3), 665 nm (B4), 842 nm (B8), • six bands at 20 metres spatial resolution– 705 nm (B5), 740 nm (B6), 783 nm (B7), 865 nm (B8A), 1 610 nm (B11), 2 190 nm (B12), • three bands at 60 metres spatial resolution: – 443 nm (B1), 945 nm (B9) and 1 375 nm (B10).
4
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 1 – data collection c) administrative data sources: • cadastral parcels vector data from Land Parcel Identification System (Agency for Restructuring and Modernisation of Agriculture ) – for image segmentation suport,
• agricultural plots borders from General Geographic Geodatabase (Head Office of Geodesy and Cartography) – for data reduction,
• information on crops declared by farmers (ARMA) – as samples for machine learning algorithms and validation
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 1 – data collection d) in situ data geodatabase for validation (CSO)
5
6
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 1 – data collection f) lists of crops for recognition – 90% of all crops of Poland
Crop classes
sugar beets
buckwheat
spring barley
winter barley
corn
cereal mixes
oat
fruit trees plantations
fruit bushes plantations
spring wheat
winter wheat
spring triticale
winter triticale
spring rape
winter rape
grassland
potatoes
rye
mustard
leguminous crops
Number of classes: 20
Aggregated crop classes
sugar beets
buckwheat
spring cereals
winter cereals
corn
fruit trees plantations
fruit bushes plantations
spring rape
winter rape
grassland
potatoes
mustard
leguminous crops
Number of classes: 13
7
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 2 – data preprocessing
a) preprocessing Sentinel-1 radar data to sigma0 (backscatter
coefficient) :
• time series data frame: from 10.2017 to 09.2018,
• number of acquisitions: approx. 28,
• number of radar scenes to be processed: approx. 56 (2 scenes for one
acquisition),
• raw data volume: approx. 100GB,
• intermediate data and individual processed products: approx. 2TB,
• open source software: ESA SNAP and CNES Orfeo Toolbox.
8
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 2 – data preprocessing
example of SENTINEL-1 data processing workflow
9
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 2 – data preprocessing
Example of processed Sentinel-1 data (false colour composition at VV polarization for 3 dates 20.05.2017, 25.06.2017, 19.07.2017, resolution 10x10m)
10
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 2 – data preprocessing
Example of processed Sentinel-1 data (false colour composition at VV polarization for 3 dates 20.05.2017, 25.06.2017, 19.07.2017, spatial resolution 10x10m)
11
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 2 – data preprocessing
a) preprocessing Sentinel-2 optical data:
• time series data frame: from 10.2017 to 08.2018,
• number of acquisitions: only 6 (due to weather conditions) ,
• number of tiles to be processed: approx. 60 (10 tiles for one
acquisition),
• raw data volume: approx. 60GB,
• intermediate data and individual processed products: approx. 200GB,
• open source software: ESA SNAP and CNES Orfeo Toolbox.
example of SENTINEL-2 data processing workflow
12
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 2 – data preprocessing
Example of processed Sentinel-2 data (natural colour composition at 28.05.2017, spatial resolution 10x10m)
13
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 2 – data preprocessing
Example of processed Sentinel-2 data (natural colour composition at 28.05.2017, spatial resolution 10x10m)
14
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 3 – testing segmentation and classification alghoritms a) Segmentation - process of partitioning a digital image into multiple homogeneous segments Example of SENTINEL-1 image segmentation • large scale mean shift segmentation algorithm will be applied for calculating homogeneous areas on Sentinel-1 and Sentinel-2 preprocessed data, • several parameters need to be tested in the context of obtaining the highest quality results for object based image classification
15
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 3 – testing segmentation and classification alghoritms b) Object Based Supervised Image Classification – several machine learning algorithms will be tested in the context of obtaining the best accuracy for crop recognition: • SVM - support vector machine classifier, • DT - decision tree classifier, • ANN - artificial neural network classifier, • RF - random forest classifier, • KNN - k-nearest neighbors classifier. For supervision and learning crops declared by farmers for direct payments in agriculture (ARMA) will be used. Database of approx. 1,5 mln records must be processed, filtered and georeferenced to obtain reliable samples.
16
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Stage 4 – accuracy assesment a) extraction of validation samples from ARMA and insitu data, b) computing confusion matrix for each classifier with ARMA and insitu validation data including: • producer accuracy per class, • user accuracy per class, • F-score measure per class, • OA - overall accuracy, • KIA - kappa index of agreement. c) comparing classified areas to the official statistics.
17
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Deliverables a) detailed description of methodology for Sentinel-1 and Sentinel-2 data processing at each stage - detailed description of methodology for Object Based Image Classification including segmentation and machine learning algorithms parameters, c) comparison of applied algorithms and their accuracy the context of obtaining best classification results, d) Crops area calculation and development of crops maps for the warmian-masurian voivodeship in 2018.
18
Case study 1 (PL) Agriculture - Crop recognition, mapping and monitoring
Deliverables
Crops map - example of Object Based Image Classification result
spring barley winter barley corn cereal mixes
oat spring wheat
winter wheat spring triticale winter triticale winter rape rye
05 – 06 December 2018, Vienna, Austria
Marek Morze
19
top related