Figure 7 - uploaded by Xiaoxiao Li
Content may be subject to copyright.
A comparison between pixel-based and object-based classification results. The first row shows the image of original DOQQ with RGB band combination display. The second row of images is the pixel-based classification results. The third row of image is the object-based classification results. Each column shows the same location on the map.
Source publication
There are growing demands for detailed and accurate land cover maps in land system research and planning. Macro-scale land cover maps normally cannot satisfy the studies that require detailed land cover maps at micro scales. In the meantime, applying conventional pixel-based classification methods in classifying high-resolution aerial imagery is in...
Context in source publication
Context 1
... OBIA method provided a robust classification result for the fine resolution land-cover mapping, and it can be used to facilitate a large amount of researches and managements in terms of the landscape planning, regional land-use and land-cover changes, environmental and sustainability as noted in the beginning of this paper. By using OBIA classification method for the land-cover mapping in our study, we identified three representative areas to demonstrate the benefit of using object-based methods (Figure 7, tables 5 and 6). In the pixel-based classification results, "salt and pepper" effects were apparent. ...
Similar publications
Object-based image analysis (OBIA) is an image classification that is oriented to object patterns that use image objects as the basis for processing, calculates characteristics per object, and extracts land cover information from remotely sensed images. This study aims to detect salt ponds using Sentinel 2 satellite data with an object-based classi...
Land use and land cover (LULC) change monitoring is still an important remote sensing application and research field, for it is closely related to manage natural resources and environmental planning. This study is aimed to find the land use and land cover changes of Zigui County from 2008 to 2014 by using multi-source remotely sensed imagery. Image...
Extraction of hydrological information from Digital Elevation Models (DEMs) is a required step when conducting any spatially distributed hydrological modelling. In particular, automated methods are proposed to extract the drainage structure from the DEM. However, a realistic river network is not always derived from conventional DEM processing metho...
Quantification of geomorphometric features is the keystone concern of the current study. The quantification was based on the statistical approach in term of multivariate analysis of local topographic features. The implemented algorithm utilizes the Digital Elevation Model (DEM) to categorize and extract the geomorphometric features embedded in the...
GEOBIA is an alternative to create and update land cover maps. In this work we assessed the combination of geographic datasets of the Cajas National Park (Ecuador) to detect which is the appropriate dataset-algorithm combination for the classification tasks in the Ecuadorian Andean region. The datasets included high resolution data as photogrammetr...
Citations
... In addition, this approach makes use of object-based classification, which involves image segmentation using spatial, spectral, and size information. This approach more closely mimics real-world features and yields more accurate categorization results for high-resolution imageries (Blaschke et al. 2014;Li and Shao 2014;Peña et al. 2014). ...
Urbanization, changes in land use and land cover (LULC), and an increase in population collectively have significant impacts on urban catchments. However, a vast majority of LULC studies have been conducted using readily available satellite imagery, which often presents limitations due to its coarse spatial resolution. Such imagery fails to accurately depict the surface characteristics and diverse spectrum of LULC classifications contained within a single pixel. This study focused on the highly urbanized Dry Creek catchment in Adelaide, South Australia and aimed to determine the impact of urbanization on spatiotemporal changes in LULC and its implications for the land surface condition of the catchment. Very high spatial resolution imagery was utilized to examine changes in LULC over the past four decades. Support Vector Machine-learning-based image classification was utilized to classify and identify the changes in LULC over the study area. The classification accuracy showed strong agreement, with a kappa value greater than 0.8. The findings of this analysis showed that extensive urban development, which expanded the built-up area by 34 km 2 , were responsible for the decline in grass cover by 43.1 km 2 over the last 40 years (1979-2019). Moreover, built-up areas, plantation, and water features, in contrast to grass cover, have demonstrated an increasing trend during the study period. The overall urban expansion over the study period was 136.6%. Urbanization intensified impervious area coverage, increasing the runoff coefficient, equivalent impervious area, and curve number by 60.6%, 60.6%, and 7.9%, respectively, while decreasing the retention capacity by 38.6%. These modifications suggest a potential variability in catchment surface runoff, prompting the need for further research to understand the surface runoff changes brought by the changes in LULC resulting from urbanization. The findings of this study can be used for land use planning and flood management.
... However, aerial images might 305 underestimate the increase in tree and shrub cover sensed by the LOVE-based estimates. Mapping 306 historical mountain vegetation at a high spatial resolution is challenging as the classification of aerial 307 images is often hindered by i) the lack of spectral discernibility between vegetation types, ii) the 308 mosaic, iii) the fact that a given vegetation formation may have a different phenology due to 310 seasonal or composite classes of vegetation; iv) the shadow effect from nearby trees or cliffs (Cots-311 Folch et al., 2007;Dirnböck et al., 2003;Dobrowski et al., 2008;Li and Shao, 2014;Zhang Id et al., 312 2020). The overall accuracy of the 2008 classified map was assessed at 82.5% when comparing the 313 classified land-cover types with their corresponding ground truth data (Haunold, 2015). ...
Quantitative estimates of past vegetation with a clearly defined spatial scale are a critical step to better understand the patterns and processes of changes in vegetation and land-cover over time and space. However, such estimates are difficult to obtain from pollen data due to their lack of spatial dimension and the inter-taxonomic differences in pollen production, dispersal, and deposition mechanisms. This study assesses the potential of the Landscape Reconstruction Algorithm (LRA) to reconstruct the spatial and temporal patterns of upland vegetation in the Bassiès Valley, northern Pyrenees over the last 200 years, a period characterised by a decline of traditional pastoral activities. We used well-dated pollen records from eight bogs and ponds and vegetation and land-cover data within a 1-km radius from each site to evaluate the LRA-based reconstruction of local land-cover. The LRA approach was then used to reconstruct the 200-year history of local land-cover dynamics at 10 to 20-year intervals around each site in relation to land-abandonment. Our study shows that in addition to allowing long-term reconstruction, LRA estimates (1) are more sensitive to vegetation composition changes over time and space compared to historical land-cover maps and pollen percentages, and (2) improve the reconstruction of past local vegetation compared to pollen percentages alone. The LRA results reveal site-specific land-cover trajectories within the small-sized study area over 200 years. Some sites exhibit changes in land-cover composition with heathland and tree cover increase (ESC, EM), while others are found to be more stable, dominated 1) exclusively by heathland (LEG, W1652), 2) heathland with intermediate proportions of grassland (FOUZ, SIG), or 3) exclusively by grassland (OT). This land-cover composition variability over space and time may be related both to local topography and past grazing activity. The quantification of past vegetation dynamics provides opportunities for bridging the gap between researchers, policy makers and practitioners, and should be taken into account to further investigate the effect of long-term grazing activities on the past and modern plant and landscape diversity.
... At the second stage of LULC mapping, we performed the segmentation and classification of the aerial photographs using an object-based approach for generating the 1955 LULC map. The object-based image analysis (OBIA) approach in LULC mapping provides advantages over the traditional per-pixel techniques such as higher classification accuracy, depicting more accurate LULC change, and differentiating extra LULC classes 33,43,44 . We used the eCognition® software (Trimble Germany GmbH, Munich) to implement an object-based image analysis (OBIA). ...
Land surface of the Earth has been changing as a result of human induced activities and natural processes. Accurate representation of landscape characteristics and precise determination of spatio-temporal changes provide valuable inputs for environmental models, landscape and urban planning, and historical land cover change analysis. This study aims to determine historical land use and land cover (LULC) changes using multi-modal geospatial data, which are the cadastral maps produced in 1858, monochrome aerial photographs obtained in 1955, and multi-spectral WorldView-3 satellite images of 2020. We investigated two pilot regions, Aksu and Kestel towns in Bursa/Turkey, to analyze the long-term LULC changes quantitatively and to understand the driving forces that caused the changes. We propose methods to facilitate the preparation of historical datasets for the LULC change detection and present an object-oriented joint classification scheme for multi-source datasets to accurately map the spatio-temporal changes. Our approach minimized the amount of manual digitizing required for the boundary delineation of LULC classes from historical geospatial data. Also, our quantitative analysis of LULC maps indicates diverging developments for the selected locations in the long period of 162 years. We observed rural depopulation and gradual afforestation in Aksu; whereas, agricultural land abandonment and deforestation in Kestel.
... Circumventing these issues requires going beyond the sole use of spectral information and including geometric elements of the urban class appearance such as pattern, shape, size, context, and orientation. Nonetheless, pixel-based classifications still fail to satisfy the accuracy requirements because they are affected by the salt-and-pepper effect and cannot fully exploit the rich information content of VHR data (Myint et al., 2011;Li and Shao, 2014). GEographic Object-Based Imagery Analysis (GEOBIA) is an alternative image processing approach that seeks to group pixels into meaningful objects based on specified parameters (Blaschke et al., 2014). ...
... There has been a quick uptake of the approach in the remote sensing community and various solutions based on deep learning have been presented recently (e.g. Sherrah, 2016;Audebert et al., 2016Audebert et al., , 2017Audebert et al., , 2018Längkvist et al., 2016;Li et al., 2015;Li and Shao, 2014;Volpi and Tuia, 2017;Liu et al., 2018;Liu et al., 2017bLiu et al., , 2018aLiu et al., , 2017aPan et al., 2018b;Marmanis et al., 2016Marmanis et al., , 2018Wen et al., 2017;Zhao et al., 2017b). A comprehensive review of deep learning applications in the field of remote sensing can be found in , Ma et al. (2019), Gu et al. (2019). ...
[free access to the paper here: https://authors.elsevier.com/a/1acEg3I9x1cfrr] Scene understanding of high resolution aerial images is of great importance for the task of automated monitoring in various remote sensing applications. Due to the large within-class and small between-class variance in pixel values of objects of interest, this remains a challenging task. In recent years, deep convolutional neural networks have started being used in remote sensing applications and demonstrate state of the art performance for pixel level classification of objects. Here we propose a reliable framework for performant results for the task of semantic segmentation of monotemporal very high resolution aerial images. Our framework consists of a novel deep learning architecture, ResUNet-a, and a novel loss function based on the Dice loss. ResUNet-a uses a UNet encoder/decoder backbone, in combination with residual connections, atrous convolutions, pyramid scene parsing pooling and multi-tasking inference. ResUNet-a infers sequentially the boundary of the objects, the distance transform of the segmentation mask, the segmentation mask and a colored reconstruction of the input. Each of the tasks is conditioned on the inference of the previous ones, thus establishing a conditioned relationship between the various tasks, as this is described through the architecture's computation graph. We analyse the performance of several flavours of the Generalized Dice loss for semantic segmentation, and we introduce a novel variant loss function for semantic segmentation of objects that has excellent convergence properties and behaves well even under the presence of highly imbalanced classes. The performance of our modeling framework is evaluated on the ISPRS 2D Potsdam dataset. Results show state-of-the-art performance with an average F1 score of 92.9% over all classes for our best model.
... This can minimize the optimal separating image into various land use classes and affect the accuracy agreement. Beyond the conventional method classifier like MLC, Li and Shao (2014) presented the object-based approach in detecting land changes at Tippecanoe County in Indiana State, USA from 1 m resolution imagery with 5-feet DEM. The study used multi-threshold (MT) segmentation to classify the land use classes. ...
... As a result, it could present three uncorrelated features that simply distinguish vegetation cover and impervious surface contained in RGB bands. Besides, the 0.5m higher resolution data were used to perform high-detection on-screen and recognize individual pixels so as to increase the chance to have the finest object segmentation and processing image classification successfully (Li & Shao, 2014). ...
Urban green space (UGS) in a city is the foundation of natural productivity in an urban structure. It is also known as a natural cooling device that plays a vital role in the city as an urban lung, discharging oxygen to reduce the city heat and as a wall against harmful air pollution. When urbanization happens, UGS, including the gazetted areas, is essentially converted into an artificial surface due to the population's demand for new development. Therefore, identifying its significance is a must and beneficial to explore. The purpose of this study is to identify the 10 years of UGS change patterns and analyze the UGS loss, particularly in the affected gazetted zone. The study used available aerial imagery data for 2002, 2012, and 2017, and database record of green space. The study had classified UGS by using the Support Vector Machine (SVM) algorithm. The training area was determined by visual interpretation and aided by a land use planning map as reference. The result validity was then determined by kappa coefficient value and producer accuracy. Overall, the study showed that the city had lost its UGS by about 88% and the total gain in built up area was 114%. The loss in UGS size in the city could be compared to a total of 2,843 units of football fields, transformed forever in just 10 years. The uncontrolled development and lack of advanced monitoring mechanism had negatively affected the planning structure of green space in KL. The implementation of advance technology as a new mitigation tool to monitor green space loss in the city could provide a variety of enhanced information that could assist city planners and urban designers to defend decisions in protecting these valuable UGS.
... According to the methodology of OBIA, image can be classified based on handcrafted features and low-level features extracted from segmented objects, such as spectral value, texture information and border information (Cánovas-García and Alonso-Sarría 2015b; Kavzoglu and Tonbul 2017a;Lucieer 2008;Zhou et al. 2009). Although OBIA methods avoid the salt-pepper effect and achieve higher accuracy than pixel-based methods (Blaschke 2001;Blaschke et al. 2008;Li and Shao 2014), misclassification in complex urban areas remains. This is because handcrafted and low-level features have weak generalization ability, which only consider the features within segmented objects' boundaries, while ignoring the detailed high-level features (e.g., semantic relationship between objects ( Cheng et al. 2017)). ...
Traditional classification methods, which use low-level features, have failed to gain satisfactory classification results of very high spatial resolution (VHR) remote sensing images. Even though per-pixel classification method based on convolutional neural network (CNN) (Per-pixel CNN) achieved higher accuracy with the help of high-level features, this method still has limitations. Per-superpixel classification method based on CNN (Per-superpixel CNN) overcome the limitations of per-pixel CNN, however, there are still some scale related issues in per-superpixel CNN needed to be explored and addressed. Firstly, in order to avoid the misclassification of complex land cover objects caused by scale effect, the superpixel classification method combining multi-scale CNN (Per-superpixel MCNN) method is proposed. Secondly, this paper analyzes how scale parameter of CNN impacts the classification accuracy and involves spatial statistics to pre-estimate scale parameter in per-superpixel image classification. This paper takes two VHR remote sensing images as experimental data, and employs two superpixel segmentation algorithms to classify urban and suburban land covers. The experimental results show that per-superpixel MCNN can effectively avoid misclassification in complex urban area compared with per-superpixel classification method combining single-scale CNN (Per-superpixel SCNN). Series of classification results also show that using the pre-estimated scale parameter can guarantee high classification accuracy, thus arbitrary nature of scale estimation can be avoided to some extent. Additionally, through discusses of the influence of accuracy evaluation method in CNN classification, it is stressed that random selection of ground truth validation points from study area is recommended and more responsibly other than using part of a reference dataset.
... However, the population can be indirectly estimated on the basis of the distribution of houses along with the average resident population per house, which is often provided by the local government. Hence, daytime land observation data are used to provide building distribution data [16][17][18]. The building level information and vacancy rate information are also considered in subsequent studies to improve the estimation accuracy [19]. ...
Previous studies have attempted to disaggregate census data into fine resolution with multisource remote sensing data considering the importance of fine-resolution population distribution in urban planning, environmental protection, resource allocation, and social economy. However, the lack of direct human activity information invariably restricts the accuracy of population mapping and reduces the credibility of the mapping process even when external facility distribution information is adopted. To address these problems, the present study proposed a novel population mapping method by combining International Space Station (ISS) photography nighttime light data, point of interest (POI) data, and location-based social media data. A similarity matching model, consisting of semantic and distance matching models, was established to integrate POI and social media data. Effective information was extracted from the integrated data through principal component analysis and then used along with road density information to train the random forest (RF) model. A comparison with WordPop data proved that our method can generate fine-resolution population distribution with higher accuracy ( R 2 = 0.91 ) than those of previous studies ( R 2 = 0.55 ). To illustrate the advantages of our method, we highlighted the limitations of previous methods that ignore social media data in handling residential regions with similar light intensity. We also discussed the performance of our method in adopting social media data, considering their characteristics, with different volumes and acquisition times. Results showed that social media data acquired between 19:00 and 8:00 with a volume of approximately 300,000 will help our method realize high accuracy with low computation burden. This study showed the great potential of combining social sensing data for disaggregating fine-resolution population.
... The first approach only considers spectral value or one aspect for boundary class [43]. Thus, PBA algorithms, with the exception of state-of-the-art convolution neural network (CNN) [28,[44][45][46], may result in a "salt and pepper" map when applied to very high-resolution images [47]. Due to the lack of an explicit object topology that might lead to inferior results compared to those from the human vision, PBA falls short of expectations in topographic mapping applications [11]. ...
The objective to fast-track the mapping and registration of large numbers of unrecorded land rights globally has led to the experimental application of Artificial Intelligence in the domain of land administration, and specifically the application of automated visual cognition techniques for cadastral mapping tasks. In this research, we applied and compared the ability of rule-based systems within Object-Based Image Analysis (OBIA), as opposed to human analysis, to extract visible cadastral boundaries from very high-resolution World View-2 images, in both rural and urban settings. From our experiments, machine-based techniques were able to automatically delineate a good proportion of rural parcels with explicit polygons where the correctness of the automatically extracted boundaries was 47.4% against 74.24% for humans and the completeness of 45% for the machine compared to 70.4% for humans. On the contrary, in the urban area, automatic results were counterintuitive: even though urban plots and buildings are clearly marked with visible features such as fences, roads and tacitly perceptible to eyes, automation resulted in geometrically and topologically poorly structured data. Thus, these could neither be geometrically compared with human digitisation, nor actual cadastral data from the field. The results of this study provide an updated snapshot with regards to the performance of contemporary machine-driven feature extraction techniques compared to conventional manual digitising. In our methodology, using an iterative approach of segmentation and classification, we demonstrated how to overcome the weaknesses of having undesirable segments due to intra-parcel and inter-parcel variability, when using segmentation approaches for cadastral feature delineation. We also demonstrated how we can easily implement a geometric comparison framework within the Esri’s ArcGIS software environment and firmly believe the developed methodology can be reproduced.
... Nonetheless, pixel-based classifications still fail to satisfy the accuracy requirements because they are 1 foivos.diakogiannis@data61.csiro.au affected by the salt-and-pepper effect and cannot fully exploit the rich information content of VHR data [42,32]. GEographic Object-Based Imagery Analysis (GEOBIA) is an alternative image processing approach that seeks to group pixels into meaningful objects based on specified parameters [6]. ...
... In recent years, deep learning methods and Convolutional Neural Networks (CNNs) in particular [29] have surpassed traditional methods in various computer vision tasks, such as object detection, semantic, and instance segmentation [see 50, for a comprehensive review]. There has been a quick uptake of the approach in the remote sensing community and various solutions based on deep learning have been presented recently [e.g., 1,2,3,28,30,32,56,33,34,35,46,47,37,58]. Some of the key advantages of CNN-based algorithms is that they provide end-to-end solutions, that require minimal feature engineering which offer greater generalization capabilities. ...
[github repo: https://github.com/feevos/resuneta] Scene understanding of high resolution aerial images is of great importance for the task of automated monitoring in various remote sensing applications. Due to the large within-class and small between-class variance in pixel values of objects of interest, this remains a challenging task. In recent years, deep convolutional neural networks have started being used in remote sensing applications and demonstrate state-of-the-art performance for pixel level classification of objects. Here we present a novel deep learning architecture, ResUNet-a, that combines ideas from various state-of-the-art modules used in computer vision for semantic segmentation tasks. We analyse the performance of several flavours of the Generalized Dice loss for semantic segmentation, and we introduce a novel variant loss function for semantic segmentation of objects that has better convergence properties and behaves well even under the presence of highly imbalanced classes. The performance of our modelling framework is evaluated on the ISPRS 2D Potsdam dataset. Results show state-of-the-art performance with an average F1 score of 92.1% over all classes for our best model.
... For instance, previous per-pixel approaches that classify each pixel in remote sensing data have used the spectral information in a single pixel from hyperspectral (or multi-spectral) imagery that consists of different channels with narrow frequency bands. This pixel-based classification method alone is known to produce salt-and-pepper effects due to misclassified pixels [16] and has had difficulties in dealing with the rich information from very high-resolution data [17,18]. Works that include more spatial information in the neighborhood of the pixel to be classified have been published [19,20]. ...
The interpretation of land use and land cover (LULC) is an important issue in the fields of high-resolution remote sensing (RS) image processing and land resource management. Fully training a new or existing convolutional neural network (CNN) architecture for LULC classification requires a large amount of remote sensing images. Thus, fine-tuning a pre-trained CNN for LULC detection is required. To improve the classification accuracy for high resolution remote sensing images, it is necessary to use another feature descriptor and to adopt a classifier for post-processing. A fully connected conditional random fields (FC-CRF), to use the fine-tuned CNN layers, spectral features, and fully connected pairwise potentials, is proposed for image classification of high-resolution remote sensing images. First, an existing CNN model is adopted, and the parameters of CNN are fine-tuned by training datasets. Then, the probabilities of image pixels belong to each class type are calculated. Second, we consider the spectral features and digital surface model (DSM) and combined with a support vector machine (SVM) classifier, the probabilities belong to each LULC class type are determined. Combined with the probabilities achieved by the fine-tuned CNN, new feature descriptors are built. Finally, FC-CRF are introduced to produce the classification results, whereas the unary potentials are achieved by the new feature descriptors and SVM classifier, and the pairwise potentials are achieved by the three-band RS imagery and DSM. Experimental results show that the proposed classification scheme achieves good performance when the total accuracy is about 85%.
