Object-based Classification Research Articles

Analysis of Landscape Structure and Spatıal Connectıvıty in Delta Systems, The Case of Big Meander Delta

Bu çalışmanın amacı Büyük Menderes Deltası Milli Parkı, Bafa Gölü Tabiat Parkı ve yakın çevresinde yer alan delta sistemindeki habitatların tanımlanması, peyzaj desenin analiz edilmesi ve mekansal bağlantılılık bakımından doğrusal habitatların tüm delta sisteminin mekansal bağlantılılığına katkılarının değerlendirilmesidir. Çalışmada, araştırma alanının mekansal özelliklerinin analizini gerçekleştirmek amacıyla kullanılan habitat haritası, 11 Ağustos 2017 tarihli Sentinel-2A uydu görüntüsü ile yardımcı veriler kullanılarak, nesne tabanlı sınıflandırma yöntemi ile elde edilmiştir. Peyzaj deseninin değerlendirilmesi ve mekansal bağlantılılığın ölçülmesinde yedi temel peyzaj metriği hesaplanmıştır. Çalışma alanında hakim habitat tipini Tarla ve Bahçe Tarımı oluşturmaktadır. Akarsular ve Durgun Tatlı Sular mekansal bağlantılığı en yüksek olan habitat tipleri iken, bunu Doğu Quercus Coccifera Garigleri, Sarcopoterium spinosum Friganaları ve Arbutus andrachne Garigleri izlemektedir. Bununla birlikte çalışma sonucunda, delta sistemlerinde akarsu koridorları ve tarla sınırları ile yol kenarlarında bulunan vejetasyon koridorlarının, diğer doğal ve yarı doğal habitatların mekansal bağlantılılığını arttırma potansiyeline sahip olduğu tespit edilmiştir.

Land Cover Classification of Nine Perennial Crops Using Sentinel-1 and -2 Data

Land cover mapping of intensive cropping areas facilitates an enhanced regional response to biosecurity threats and to natural disasters such as drought and flooding. Such maps also provide information for natural resource planning and analysis of the temporal and spatial trends in crop distribution and gross production. In this work, 10 meter resolution land cover maps were generated over a 6200 km2 area of the Riverina region in New South Wales (NSW), Australia, with a focus on locating the most important perennial crops in the region. The maps discriminated between 12 classes, including nine perennial crop classes. A satellite image time series (SITS) of freely available Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 multispectral imagery was used. A segmentation technique grouped spectrally similar adjacent pixels together, to enable object-based image analysis (OBIA). K-means unsupervised clustering was used to filter training points and classify some map areas, which improved supervised classification of the remaining areas. The support vector machine (SVM) supervised classifier with radial basis function (RBF) kernel gave the best results among several algorithms trialled. The accuracies of maps generated using several combinations of the multispectral and radar bands were compared to assess the relative value of each combination. An object-based post classification refinement step was developed, enabling optimization of the tradeoff between producers’ accuracy and users’ accuracy. Accuracy was assessed against randomly sampled segments, and the final map achieved an overall count-based accuracy of 84.8% and area-weighted accuracy of 90.9%. Producers’ accuracies for the perennial crop classes ranged from 78 to 100%, and users’ accuracies ranged from 63 to 100%. This work develops methods to generate detailed and large-scale maps that accurately discriminate between many perennial crops and can be updated frequently.

Dimensionality Reduction and Feature Selection for Object-Based Land Cover Classification based on Sentinel-1 and Sentinel-2 Time Series Using Google Earth Engine

Mapping Earth’s surface and its rapid changes with remotely sensed data is a crucial task to understand the impact of an increasingly urban world population on the environment. However, the impressive amount of available Earth observation data is only marginally exploited in common classifications. In this study, we use the computational power of Google Earth Engine and Google Cloud Platform to generate an oversized feature set in which we explore feature importance and analyze the influence of dimensionality reduction methods to object-based land cover classification with Support Vector Machines. We propose a methodology to extract the most relevant features and optimize an SVM classifier hyperparameters to achieve higher classification accuracy. The proposed approach is evaluated in two different urban study areas of Stockholm and Beijing. Despite different training set sizes in the two study sites, the averaged feature importance ranking showed similar results for the top-ranking features. In particular, Sentinel-2 NDVI, NDWI, and Sentinel-1 VV temporal means are the highest ranked features and the experiment results strongly indicated that the fusion of these features improved the separability between urban land cover and land use classes. Overall classification accuracies of 94% and 93% were achieved in Stockholm and Beijing study sites, respectively. The test demonstrated the viability of the methodology in a cloud-computing environment to incorporate dimensionality reduction as a key step in the land cover classification process, which we consider essential for the exploitation of the growing Earth observation big data. To encourage further research and development of reliable workflows, we share our datasets and publish the developed Google Earth Engine and Python scripts as free and open-source software.

Object-based post-fire aerial image classification for building damage, destruction and defensive actions at the 2012 Colorado Waldo Canyon Fire

We compare the use of post-fire aerial imagery to ground-based assessment for identifying building destruction and damage at the 2012 Colorado Waldo Canyon Fire. We also compare active-fire defensive actions identified via manual and automated post-fire image classification to defensive actions documented from ground-based assessments (witness discussions, vehicle logs and images). For building destruction, manual and automatic image classification compared favourably to ground-based assessment, with low errors of commission (0.0–0.4%) and omission (0–1.2%). For building damage, classifying imagery manually had significant errors of commission and omission (59.0% and 57.9%) because ground-based assessments missed roof damage and image classification excluded interior and side damage, indicating the need for both techniques. Classifying imagery automatically for indicators of active-fire water suppression on buildings has Kappa statistics indicating a substantial agreement with documented water suppression. Manual and automatic image classification underestimated the full extent of documented defensive actions but showed a statistically significant dependence between fire cessation and defensive actions. These results show post-fire imagery to be a useful addition to other techniques for identifying building damage, destruction and defensive actions. Also demonstrated is the importance of accounting for defensive actions and other factors in wildland–urban interface fire studies.

Land Cover Classification of Complex Agroecosystems in the Non-Protected Highlands of the Galapagos Islands

The humid highlands of the Galapagos are the islands’ most biologically productive regions and a key habitat for endemic animal and plant species. These areas are crucial for the region’s food security and for the control of invasive plants, but little is known about the spatial distribution of its land cover. We generated a baseline high-resolution land cover map of the agricultural zones and their surrounding protected areas. We combined the high spatial resolution of PlanetScope images with the high spectral resolution of Sentinel-2 images in an object-based classification using a RandomForest algorithm. We used images collected with an unmanned aerial vehicle (UAV) to verify and validate our classified map. Despite the astounding diversity and heterogeneity of the highland landscape, our classification yielded useful results (overall Kappa: 0.7, R2: 0.69) and revealed that across all four inhabited islands, invasive plants cover the largest fraction (28.5%) of the agricultural area, followed by pastures (22.3%), native vegetation (18.6%), food crops (18.3%), and mixed forest and pioneer plants (11.6%). Our results are consistent with historical trajectories of colonization and abandonment of the highlands. The produced dataset is designed to suit the needs of practitioners of both conservation and agriculture and aims to foster collaboration between the two areas.

Identifying and Quantifying the Abundance of Economically Important Palms in Tropical Moist Forest Using UAV Imagery

Sustainable management of non-timber forest products such as palm fruits is crucial for the long-term conservation of intact forest. A major limitation to expanding sustainable management of palms has been the need for precise information about the resources at scales of tens to hundreds of hectares, while typical ground-based surveys only sample small areas. In recent years, small unmanned aerial vehicles (UAVs) have become an important tool for mapping forest areas as they are cheap and easy to transport, and they provide high spatial resolution imagery of remote areas. We developed an object-based classification workflow for RGB UAV imagery which aims to identify and delineate palm tree crowns in the tropical rainforest by combining image processing and GIS functionalities using color and textural information in an integrative way to show one of the potential uses of UAVs in tropical forests. Ten permanent forest plots with 1170 reference palm trees were assessed from October to December 2017. The results indicate that palm tree crowns could be clearly identified and, in some cases, quantified following the workflow. The best results were obtained using the random forest classifier with an 85% overall accuracy and 0.82 kappa index.

Comparison of Pixel-Based and Object-Based Classification Methods in Determination of Wetland Coastline

Comparison of Pixel-Based and Object-Based Classification Methods in Determination of Wetland Coastline

An Object-Based Hierarchical Compound Classification Method for Change Detection in Heterogeneous Optical and SAR Images

Change detection in heterogeneous remote sensing images is an important but challenging task because of the incommensurable appearances of the heterogeneous images. In order to solve the change detection problem in optical and synthetic aperture radar (SAR) images, this paper proposes an improved method that combines cooperative multitemporal segmentation and hierarchical compound classification (CMS-HCC) based on our previous work. Considering the large radiometric and geometric differences between heterogeneous images, first, a cooperative multitemporal segmentation method is introduced to generate multi-scale segmentation results. This method segments two images together by associating the information from the two images and thus reduces the noises and errors caused by area transition and object misalignment, as well as makes the boundaries of detected objects described more accurately. Then, a region-based multitemporal hierarchical Markov random field (RMH-MRF) model is defined to combine spatial, temporal, and multi-level information. With the RMH-MRF model, a hierarchical compound classification method is performed by identifying the optimal configuration of labels with a region-based marginal posterior mode estimation, further improving the change detection accuracy. The changes can be determined if the labels assigned to each pair of parcels are different, obtaining multi-scale change maps. Experimental validation is conducted on several pairs of optical and SAR images. It consists of two parts: comparison on different multitemporal segmentation methods and comparison on different change detection methods. The results show that the proposed method can effectively detect the changes in heterogeneous images, with low false positive and high accuracy.

Land Cover Classification using Random Forest Technique and DEM Auxiliary Data

This article has been written based on findings and results of author’s master thesis. Nowadays, every month/year changes occur on the surface of the Earth from a small build ups till large-scale natural disasters that destroyed entire settlements. Observations and analysis using satellite data can help to improve and prevent any natural or anthropogenic disaster, plan and calculate beneficial usage of the land cover. Remote Sensing and Spatial Analysis tools are one of the approaches for them. This study tries to sustain one of the advantageous approaches for extracting information from semi-automatic mapping. Besides, article provides semi-automatic methodology of extracting information from multispectral image for large areas. For that, the Sentinel-2a (S2) optical data with 10 m resolution for Land Cover Classification System (LCCS) classification and as auxiliary data to extract the urban and industrial areas DEM and build up indices were used. Object-based image classification with random forest classification techniques using LCCS classes was done with an accuracy of 88%. Moreover Salzkammergut was chosen as a study area. The Salzkammergut is a historical and recreational area part of Austria, consisting of 52 districts from three federal states and with no official administrative boundaries.

Temporal stage and method selection of tree species classification based on GF-2 remote sensing image

It's important to master tree species composition and distribution in forests for the study of forest ecosystems. To promote the application of domestic Gaofen data in the classification of tree species and to explore the effects of different combining images, classification features and classifier on tree species classification results, three kinds of single temporal data and four kinds of multi-temporal data were constructed. Based on three GF-2 images, according to the multi-scale segmentation, C5.0 feature optimization as well as two classifiers including support vector machine (SVM) and random forest (RF), we finished the object-based classification of eight tree species of different temporal and feature dimensions respectively, and finally achieved good results with overall accuracy between 63.5% and 83.5% and the Kappa coefficient between 0.57 and 0.81. The results showed that the choice of temporal stage would affect the classification results. The results based on multi-temporal were generally better than that on single temporal stage. There were obvious precision differences between different image combinations of multi-temporal as well as different single temporal stage. It is notable that feature optimization played a positive role in the improvement of classification accuracy. SVM was relatively stable across different temporal and feature dimensions, which should be given priority when single temporal and classification features are difficult to distinguish tree species directly, while it should be noted that SVM was easy to overfit. RF was not easy to overfit, but it was more dependent on the quality of classification features and would get good results under favorable image combination.

Mapping of Substrate Categories Changes in Exposure to Coral Reefs Using Alos Avnir-2 and Spot 7 Images in The South Area of Kotabaru District

The South of Pulau Laut is an area in the which is included in the Coral Triangle and Aquatic Conservation Areas. The purpose of this study was to map changes in the extent of coral reefs based on the class of substrate categories and objects on exposure to coral reefs in the southern waters of Sea Island within 6 years from 2011- 2017 using SPOT 7 satellite imagery in 2017 and AVNIR ALOS-2 in 2011. Using the method of Lyzenga Dept Index for basic invariant detection of waters. Based on the results of substrate-based classification using both satellite images overlaye, there has been a decline for the coral category as much as -19.21 ha, seagrass and algae as much as -3.57 ha. On the Contrary, the category of open substrate has Increased by 22.77 ha. Object-based classification results using the Object Based Image Analysis (OBIA) method have a reduction in area for the hard coral category (HC) of -1.66 ha, hard sand mixed with sand (HC S) of -15.99 ha, dead coral overgrown with algae and sand (DCA S) of -0.39 ha, algae and seagrass (FS) covering an area of ​​-1.65 ha, sand-mixed coral fragments (RS) of 4:45 ha. On the other hand, there was an increase of in the area.

Object-based feature extraction for hyperspectral data using firefly algorithm

Object-based classification methods can improve the accuracy of hyperspectral image classification due to the fact that they incorporate spatial information into the classification procedure by assigning neighboring pixels into the same class. In this paper, a new object-based feature extraction method is proposed which makes use of information theory to reduce the Bayes error. In this way, the proposed method exploits higher order statistics for feature extraction which are very effective for non Gaussian data such as hyperspectral images. The criterion to be minimized is composed of three mutual information terms. The first and second terms, consider the maximal relevance and minimal redundancy, respectively, while the third term takes into account the segmentation map containing disjoint spatial regions. To obtain the segmentation map, we apply the firefly clustering algorithm whose fitness function simultaneously considers the intra-distance between samples and their cluster centroids, and inter-distance between centroids of any two clusters. Our experimental results, performed using a variety of hyperspectral scenes, indicate that the proposed framework gives better classification results than some state-of the-art spectral–spatial feature extraction methods.

Land Cover/Land Use Mapping of LISS IV Imagery Using Object-Based Convolutional Neural Network with Deep Features

The paper proposes a new method for classifying the LISS IV satellite images using deep learning method. Deep learning method is to automatically extract many features without any human intervention. The classification accuracy through deep learning is still improved by including object-based segmentation. The object-based deep feature learning method using CNN is used to accurately classify the remotely sensed images. The method is designed with the technique of extracting the deep features and using it for object-based classification. The proposed system extracts deep features using pre-defined filter values, thus increasing the overall performance of the process compared to randomly initialized filter values. The object-based classification method can preserve edge information in complex satellite images. To improve the classification accuracy and to reduce complexity, object-based deep learning technique is used. The proposed object-based deep learning approach is used to drastically increase the classification accuracy. Here, the remotely sensed images were used to classify the urban areas of Ahmadabad and Madurai cities. Experimental results show a better performance with the object-based classification.

Glacier Mapping with Object based Image Analysis Method, Case Study of Mount Everest Region

Substantial progress in Geoinformatics System in recent years leads to the research in monitoring and mapping of glaciers. Monitoring glacier in the mountain region with traditional manual method is very crucial and time-consuming. Glaciers are melting because of global warming. Melting of glaciers can causes calamities like rising in sea level, glacial lake outburst, avalanches etc. Glacier monitoring using multi-temporal data for objects on the surface of the glacier is hard to classify. This paper gives an insight into the importance of Geo-spatial data and object-based image analysis method for satellite image processing. The object-based image analysis benefits more compared to traditional pixel-based image analysis as it is more robust and noise removing more image features etc. Spectral data with multiple bands is the backbone of surveying and monitoring of glacier. In this paper case study of Mount Everest region (27 48° 32N, 86 54° 47E) is represented. The remotely sensed image needs to be taken in a cloud-free environment. Object-based image classification is done in recognition tool. Also, the step by step methodology of object-based classification, segmentation and post-classification possibilities are discussed. Finally, the paper presents several representations of indexes. The integration of indexes is useful for accurately classifying the different part of terrain, lake, vegetation and glacier.

Object based classification of high resolution remote sensing image using HRSVM-CNN classifier

ABSTRACT Remote Sensing (RS) implies the science of recognition of earth surface features utilizing electromagnetic radiation. A lot of images encompass a higher spatial and also radiometric resolution. To obtain the required information from the various satellite images, efficient classification and also image processing is needed. Also, in existent works, the accuracy is low. To overcome such drawbacks, this paper proposed an object-based classification (OBC) of high-resolutions RS image (HRRSI) using HRSVM-CNN classifier. Initially, the input image is improved utilizing hybrid CLAHE-DCT in the preprocessing stage. The improved image is then segmented by using MRBIS algorithm. The MRBIS segmentation has five steps, such as seed selection using seed generation, FCM, region growing (RG), segment initialization, and also region merging. In the RG step, the image is segmented as regions centered on space, spectral, location, and area. Subsequently, the LTrP, color histogram, GLCM, GLDM, edge feature, shape features are extorted as of the segmented image. Finally, the classifier named as HRSVM-CNN is utilized to classify the input as four classes, for instance, Natural Vegetation, Road, Residential, and Water-bodies. The performance proffered by the proposed segmentation and classification techniques is contrasted to the prevailing techniques centered on statistical measures.

Quantitative assessment of urbanization and impacts in the complex of Huế Monuments, Vietnam

Rapid and extensive urban changes in recent decades have inflicted a multitude of challenges in land-use planning and conservation management, especially for the heritage protection of historic areas. This paper describes the results of an integrated approach to analyzing the potential environmental impacts of urbanization that has profoundly transformed land cover and land -use in Vietnam. A combination of remote-sensing techniques and ancillary indices enables a quantitative evaluation of urbanization impacts on 12 sites in the Complex of Huế Monuments, a UNESCO World Heritage Site, from 1968 to 2016. This approach includes classifying satellite images, analyzing urbanization characteristics with land-cover and land-use change and urbanization indices, and quantifying environmental impacts by ecosystem service values. Object-based classification results indicate that the support vector machine algorithm achieves an optimum overall accuracy of 84.78–86.5% and a Kappa coefficient of 0.82–0.84. Quantitative results of the total absolute loss value reveal the largest negative effect at Huế Citadel and the most positive at Tự Đức Tomb, with the overall consequence of urbanization being predominantly negative for the whole complex. The adverse impacts are found primarily from the degradation of agriculture and green space. The negative total loss value also serves as a quantification of impacts from increased urbanization on human well-being. This integrated methodology can be a potentially effective tool to plan for sustainable development in the continual trend toward further urbanization in Vietnam.

The contribution of ALOS/PALSAR-1 multi-temporal data to map permanently and temporarily flooded coastal wetlands

ABSTRACTAlthough regional wetland mapping studies have mostly relied on optical sensors, synthetic aperture radar (SAR) sensors are being increasingly applied. The aim of this study is to analyse the ability of the Phased Array type L-band Synthetic Aperture Radar on board of the Advanced Land Observing Satellite (ALOS/PALSAR-1) data to identify, delineate and monitor wetlands, and to evaluate the importance of scene selection in a highly unpredictable wetland. Three SAR scene sets (Year A, Year B and Inter-annual) were built for this purpose, considering the intra-annual and inter-annual hydrologic variability and the phenologic variability of the studied coastal wetland. Seven land cover types were defined, including three permanently flooded wetland classes, three temporarily flooded wetland classes and one non-wetland class. An object-based unsupervised classification approach was applied on each multi-temporal set. The obtained clusters were characterized by a temporal signature and assigned to the seven land cover types using a decision tree with user-defined thresholds. The accuracy assessment of each product was performed using a set of 258 data sites, including field collected data and data retrieved from Landsat 8 Operational Land Imager (OLI) imagery acquired during the dates of the field campaign. The Year B set showed the best accuracy (83.4% overall, 75% Kappa coefficient (κ)) and the lowest omission and commission mean errors (16.6% and 16.1% respectively). The classes that were best differentiated are permanently flooded wetlands (PFW) and non-wetlands (NW) in all sets.

Geo-parcel-based crop classification in very-high-resolution images via hierarchical perception

ABSTRACTThe basic application of remote sensing is classifying surface objects in images. Traditional pixel-based or object-based classification methods are poorly suited to very high-resolution (VHR) images captured by remote sensors with high spatial resolutions. In the field of computer vision, deep learning has recently achieved great advances in natural image processing. Inspired by this, we propose a methodology guided by hierarchical perception to classify crops in VHR images based on geo-parcels. Geo-parcel-based crop classification is used in agriculture and in refined farmland management. The proposed methodology can be divided into three steps: zoning, location and quality. In the first step, the image is divided into blocks based on the road network. In the second step, geographical entities are extracted from every block defined in the zoning step. In the last step, the geographical entity types are identified based on the texture information. These steps provide mutual constraints. In each step, the information is extracted by neural networks that have been adapted to the VHR images. The experimental results indicate that our methodology performs well, with a precision greater than 90%. Furthermore, our methodology combines deep learning techniques and theory regarding image perception by humans, providing a valuable method for processing remote sensing information.

LAND USE/LAND COVER ASSESSMENT IN A SEISMICALLY ACTIVE REGION IN KUNDASANG, SABAH

Abstract. Land use development in the mountainous environment must be risk-informed especially in the area highly vulnerable to disaster and extreme climate. Kundasang, Sabah, Malaysia is one of the tourist-demanding areas characterized by mountainous landscape and agriculture activity. The increasing number of tourists and agriculture activity affects the land use exploration. This area is vulnerable to geohazards such as earthquakes and landslides due to its location under seismically active region and complex geological environment. In this study, geospatial technique was used to assess the land use activity in Kundasang, Sabah pertaining to geological risk in this area. The assessment started with the identification of geohazard activities and its associated tectonic features using field investigation and mapping for coherent visualization. Subsequently, multiple high-resolution satellite imageries were used to detect land use changes before and after the disaster. In order to detect the land use change, object-based change detection was applied based on segmentation and object-based classification compared to the classical pixel-based method. The output of this study shows a number of field evidences associated with geohazard features that affecting the land use activities especially build-up area and agriculture land. In conclusion, the combined results provide an important benefit for better understanding the interaction between geohazard activity and landscape patterns in order to support the planning and decision making through spatial analysis and appropriate object-based processing method.

Su Kütlesi Belirlemede Farklı Sınıflandırma Yöntemlerinin Karşılaştırılması: Atikhisar Barajı (Çanakkale) Örneği

Water resources management is one of the most important issues of today. Satellite remote sensing have been successfully used to detect the presence of water bodies. In this study, four remote sensing methods: (1) normalized difference water index (NDWI), (2) support vector machine (SVM), (3) geographic object-based image analysis (GEOBIA) and (4) NDWI supported GEOBIA (GEOBIA_NDWI) were examined for water body area detection. For this purpose, Atikhisar Dam Lake, the only water source of Çanakkale central district of Turkey was selected as study area. As remote sensing data nine multitemporal Landsat-8 Operational Land Imager (OLI) multispectral satellite images between 2013 and 2017 were used. For the accuracy assessment, area values extracted from the used methods were tested with in-situ measurement lake area values. The main issues discussed in this study can be specified as follows: (i) Is pixel-based classification SVM or object-based image classification GEOBIA more successful in the water body detection?, (ii) Are the image classification methods (SVM and GEOBIA) or the water index (NDWI) more successful in the water body detection? and (iii) What is the contribution of NDWI to GEOBIA_NDWI (GEOBIA_NDWI) classification in the water body detection? The results show that meteorological factors and irrigation were influential in lake area variations. NDWI was found to be superior to other methods in determining water body and allowed for better detection of the lake boundary. Additionally, NDWI made a better separation of the land cover classes adjacent to water at the border. The object based GEOBIA was better than the pixel based SVM for distinguishing water and other land cover classes adjacent to border. GEOBIA_NDWI lake area results were more accurate than the standard object-based classification. Mixed pixels out of the lake area was determined less in the NDWI and GEOBIA_NDWI results.