HIERARCHICAL CLASSIFICATION OF DIFFERENT CROPS USING

Slides:



Advertisements
Similar presentations
GLC 2000 ‘Final Results’ Workshop (JRC-Ispra, 24 ~ 26 March, 2003) GLC 2000 ‘Final Results’ Workshop (JRC-Ispra, 24 ~ 26 March, 2003) LAND COVER MAP OF.
Advertisements

September 5, 2013 Tyler Jones Research Assistant Dept. of Geology & Geography Auburn University.
Utilization of Remotely Sensed Data for Targeting and Evaluating Implementation of Best Management Practices within the Wister Lake Watershed, Oklahoma.
Major Operations of Digital Image Processing (DIP) Image Quality Assessment Radiometric Correction Geometric Correction Image Classification Introduction.
Use of Remote Sensing and GIS in Agriculture and Related Disciplines
Application Of Remote Sensing & GIS for Effective Agricultural Management By Dr Jibanananda Roy Consultant, SkyMap Global.
Multiple Criteria for Evaluating Land Cover Classification Algorithms Summary of a paper by R.S. DeFries and Jonathan Cheung-Wai Chan April, 2000 Remote.
1 Learning to Detect Objects in Images via a Sparse, Part-Based Representation S. Agarwal, A. Awan and D. Roth IEEE Transactions on Pattern Analysis and.
GDA Corp. GIS Expert System for Riparian Buffer Delineation and LC mapping Riparian Buffer GIS Meeting Annapolis, MD February 6, 2007 Dmitry L. Varlyguin.
VALIDATION OF REMOTE SENSING CLASSIFICATIONS: a case of Balans classification Markus Törmä.
Lecture 14: Classification Thursday 18 February 2010 Reading: Ch – 7.19 Last lecture: Spectral Mixture Analysis.
Digital Imaging and Remote Sensing Laboratory Real-World Stepwise Spectral Unmixing Daniel Newland Dr. John Schott Digital Imaging and Remote Sensing Laboratory.
The use of Remote Sensing in Land Cover Mapping and Change Detection in Somalia Simon Mumuli Oduori, Ronald Vargas Rojas, Ambrose Oroda and Christian Omuto.
INTRODUCTION Problem: Damage condition of residential areas are more concerned than that of natural areas in post-hurricane damage assessment. Recognition.
Module 2.1 Monitoring activity data for forests using remote sensing REDD+ training materials by GOFC-GOLD, Wageningen University, World Bank FCPF 1 Module.
Accuracy Assessment. 2 Because it is not practical to test every pixel in the classification image, a representative sample of reference points in the.
An Object-oriented Classification Approach for Analyzing and Characterizing Urban Landscape at the Parcel Level Weiqi Zhou, Austin Troy& Morgan Grove University.
Ramesh Gautam, Jean Woods, Simon Eching, Mohammad Mostafavi, Scott Hayes, tom Hawkins, Jeff milliken Division of Statewide Integrated Water Management.
U.S. Department of the Interior U.S. Geological Survey Assessment of Conifer Health in Grand County, Colorado using Remotely Sensed Imagery Chris Cole.
Classification & Vegetation Indices
Introduction to Remote Sensing. Outline What is remote sensing? The electromagnetic spectrum (EMS) The four resolutions Image Classification Incorporation.
Operational Agriculture Monitoring System Using Remote Sensing Pei Zhiyuan Center for Remote Sensing Applacation, Ministry of Agriculture, China.
Seto, K.C., Woodcock, C.E., Song, C. Huang, X., Lu, J. and Kaufmann, R.K. (2002). Monitoring Land-Use change in the Pearl River Delta using Landsat TM.
ASPRS Annual Conference 2005, Baltimore, March Utilizing Multi-Resolution Image data vs. Pansharpened Image data for Change Detection V. Vijayaraj,
The role of remote sensing in Climate Change Mitigation and Adaptation.
Fuzzy Entropy based feature selection for classification of hyperspectral data Mahesh Pal Department of Civil Engineering National Institute of Technology.
Image Classification 영상분류
Change Detection Techniques D. LU, P. MAUSEL, E. BRONDIZIO and E. MORAN Presented by Dahl Winters Geog 577, March 6, 2007.
Understanding Glacier Characteristics in Rocky Mountains Using Remote Sensing Yang Qing.
Ramesh Gautam, Jean Woods, Simon Eching, Mohammad Mostafavi Land Use Section, Division of Statewide Integrated Water Management California Department of.
Crop Mapping in Stanislaus County using GIS and Remote Sensing Ramesh Gautam, Jean Woods, Simon Eching, Mohammad Mostafavi Land Use Section, Division of.
Map of the Great Divide Basin, Wyoming, created using a neural network and used to find likely fossil beds See:
Wetlands Investigation Utilizing GIS and Remote Sensing Technology for Lucas County, Ohio: a hybrid analysis. Nathan Torbick Spring 2003 Update on current.
Change Detection The process of identifying differences in the state of an object or phenomenon by observing it at different times. Change detection applications:
Remote Sensing Classification Accuracy
Progress Report. Project Description Purpose: The goals of this project are to build a geodatabase and create a brochure with a map atlas for the Freeman.
14 ARM Science Team Meeting, Albuquerque, NM, March 21-26, 2004 Canada Centre for Remote Sensing - Centre canadien de télédétection Geomatics Canada Natural.
BOT / GEOG / GEOL 4111 / Field data collection Visiting and characterizing representative sites Used for classification (training data), information.
earthobs.nr.no Retraining maximum likelihood classifiers using a low-rank model Arnt-Børre Salberg Norwegian Computing Center Oslo, Norway IGARSS.
Fuzzy Regions for Handling Uncertainty in Remote Sensing Image Segmentation Ivan Lizarazo, (a) and Paul Elsner (b) (a) Department of Cadastral Engineering.
Object-oriented Land Cover Classification in an Urbanizing Watershed Erik Nordman, Lindi Quackenbush, and Lee Herrington SUNY College of Environmental.
Citation: Moskal., L. M. and D. M. Styers, Land use/land cover (LULC) from high-resolution near infrared aerial imagery: costs and applications.
GLC 2000 Workshop March 2003 Land cover map of southern hemisphere Africa using SPOT-4 VEGETATION data Ana Cabral 1, Maria J.P. de Vasconcelos 1,2,
APPLIED REMOTE SENSING TECHNOLOGY TO ANALYZE THE LAND COVER/LAND USE CHANGE AT TISZA LAKE By Yudhi Gunawan * and Tamás János ** * Department of Land Use.
Mapping Canada’s Rangeland and Forage Resources using Earth Observation Emily Lindsay MSc Candidate – Carleton University Supervisors: Doug J. King & Andrew.
High Spatial Resolution Land Cover Development for the Coastal United States Eric Morris (Presenter) Chris Robinson The Baldwin Group at NOAA Office for.
Unsupervised Classification
Integrating LiDAR Intensity and Elevation Data for Terrain Characterization in a Forested Area Cheng Wang and Nancy F. Glenn IEEE GEOSCIENCE AND REMOTE.
26. Classification Accuracy Assessment
Gofamodimo Mashame*,a, Felicia Akinyemia
Pathik Thakkar University of Texas at Dallas
Factsheet # 12 Understanding multiscale dynamics of landscape change through the application of remote sensing & GIS Land use/land cover (LULC) from high-resolution.
A Personal Tour of Machine Learning and Its Applications
Environmental Intelligence Platform – Monitoring Nutrients Pollution with Earth Observation Data for Sustainable Agriculture and Clean Waters Blue.
Mapping wheat growth in dryland fields in SE Wyoming using Landsat images Matthew Thoman.
Map of the Great Divide Basin, Wyoming, created using a neural network and used to find likely fossil beds See:
National seminar on ENVIRONMENT AND DEVELOPMENT IN EASTERN INDIA
Incorporating Ancillary Data for Classification
By Yudhi Gunawan * and Tamás János **
Evaluating Land-Use Classification Methodology Using Landsat Imagery
الدكتور: أحمد رأفت غضية صفاء عبد الجليل كامل حمادة
Paulo Gonçalves1 Hugo Carrão2 André Pinheiro2 Mário Caetano2
REMOTE SENSING Multispectral Image Classification
Supervised Classification
Image Information Extraction
Igor Appel Alexander Kokhanovsky
Ensemble Methods: Bagging.
Remote Sensing Landscape Changes Before and After King Fire 2014
Calculating land use change in west linn from
An introduction to Machine Learning (ML)
Presentation transcript:

HIERARCHICAL CLASSIFICATION OF DIFFERENT CROPS USING MEDIUM AND HIGH SPATIAL RESOLUTION IMAGES Fonseca-Luengo, D., de la Fuente-Sáiz, D., Fuentes-Peñailillo, F., and Ortega-Farías, S. Centro de Investigación y Transferencia en Riego y Agroclimatología (CITRA), Universidad de Talca, Chile. *sortega@utalca.cl dfonseca@utalca.cl 1 Introduction Results and Discussion Pattern recognition for image classification is a key objective that has been followed in the remote sensing area. This is due to the importance of crop identification for tasks such as: cadastral process, management policy, decision making, crop status, landcover evolution, production and social interaction, etc. The major availability of satellite images, among other uses, has allowed for the improvement of land cover maps in order to detect changes in temporal and spatial patterns. In this sense, different classification methods have been developed, which indicated that a combination of individual classifiers, called ensemble, can generate more accurate results than individual classifiers. One of them is the Random Forest (RF) classifier, which is a combination of numerous classification trees that provides a prediction considering the most frequent individual vote by each tree. The main goal of this research was to evaluate a hierarchical classification of agricultural landcover through high and medium spatial resolution satellite images, i.e. Spot 6/7 and Landsat-7 ETM+, in an area belonging to the Ancoa basin, Maule Region, Chile. This was made considering two RF classifiers structured to generate multi-level classes in function of the input image, and using the GEOBIA (Geographic Object Based Image Analysis) approach to improve the extraction of the features from training data. The GEOBIA approach allowed to generate a greater number of samples for training process, because each polygons (manually delineated in the cadastral step) were sub- divided in a group of segments that were used then as new samples. In this sense, a fine segmentation can be helpful to generate a great number of samples. OBB error for level 4 was greater (30%) than for level 3 (10%), which can be explain due that in the level 4 the classes can be more similar between them, generating errors in the separation process. Validation of classified maps was carried out using a confusion matrixes for each experiments. Overall accuracies for all experiments ranged between 70% and 93%, comparing to those classes generated with the validation ROIs delineated in the cadastral process. Comparing maps from L3 (Figure 2 (a)) and L4 (Figure 2 (c)), the main differences can be observed between the classes: Bare Soil and Urban; and Flooded rice and Crop. These classes were misclassificated due to incorporation of other similar spectral responses from additional classes of level 4. It is important to highlight, that in both Spot maps were correctly identified the class Flooded rice, which is an abundant crop in the west-south area of the study site (and confirmed with the cadaster step). Classified map for Landsat image (Figure 2 (b)) shows less details in the class recognition compared with map from degraded Spot images (to 15 meters), despite both have same spatial resolution. which could be attributed to the pansharpening step. In addition, zones with Urban class were misclassificated as Bare soil, and Forest class was misclassified by Natural vegetated areas and Crop. In contrast, classification with Landsat image was able to identify irrigation canals, but as a Natural vegetated due to the interaction between the surfaces of water, concrete and poor vegetation. Materials and Methods Depending on kind of input, specific classes in different levels (hierarchical levels) can be generated, i.e., level 3 for Landsat and level 4 for Spot image (shown in the next figure). Study site. The study site corresponds to an agricultural area of 98,000 ha belonging to the Ancoa basin, fed by the Ancoa reservoir, located in the Maule Region in the central valley of Chile. Satellite images and image processing. Satellite images collected by Landsat 7 ETM+ and Spot 7 sensors were used. Spot image was captured on December 20 of 2015, in the beginning of the highest water demand in the study site. While Landsat image was captured on December 22 of 2015. For the image processing, the spatial resolution of multispectral bands in both satellite images were increased using a pansharpening method, aiming to improve the extraction of the features. In addition, GEOBIA approach was implemented considering segments instead of pixels in the training and implementation processes using Simple Linear Iterative Clustering (SLIC) algorithm for segmentation process. Classifier. Ground true classes on the different bands were used to train a RF model to classify different crops, considering mean and standard deviation for features in segments from: multispectral bands, normalized difference vegetation index (NDVI), Tasseled Cap transformation, and Entropy. Two RF models were generated for both levels considering 100 trees for each one, i.e., Spot image for level 4 and degraded Spot until spatial resolution of Landsat for level 3. Finally, classification with Spot and Landsat images were evaluated. Validation. Accuracy assessment was carried out considering some ground true classes that were not used in the training step. Conclusions The proposed methodology aims to classify land covers with a GEOBIA approach at different hierarchical scales considering a training step using Spot 7 image with pansharpening and degradation treatments. This approach generated a classification model composed by two classifiers for two scale levels. In addition, the implementation of this model for Landsat images made it possible to achieve stable results. Similar results were generated in both maps from Spot images (original and pansharpened), where classes at level 3 were very consistent. Regarding Landsat, some misclassifications were found in classes: Urban, Bare soil, and Forest, which could be generated by a poor incorporation of information from the pansharpening step. Since only features from cadastral process were used, it is important to point out the possible improvements of this proposal considering extractions of multi-temporal features from other scenes in different phenological stages. This future work should aim to integrate the temporal behavior of the training data set in the classification model, which would add important features in the learning model. ACKNOWLEDGEMENTS. The research leading to this report was supported by the A2C2 Program "Adapting Agriculture to Climate Change", Facultad de Ciencias Agrarias, Universidad de Talca, Talca, Chile.