Pixel-based Image Classification Research Articles

Benthic Habitat Classification Using Sentinel-2a Image With And Without Water Column Correction In Pengudang Village, Bintan Regency

Research on the mapping of benthic habitats in the waters of Pengudang Village, Bintan Regency, uses Sentinel-2A imagery. This mapping process has the potential to display the results of classification of benthic habitats in shallow waters. The purpose of this study was to assess the Maximum Likelihood Classification (MLH) algorithm model for mapping benthic habitat cover through Sentinel-2A satellite imagery using the lyzenga water column correction and without water column correction and to calculate the accuracy of the MLH classification results for benthic habitats in waters Pengudang Village. This classification uses a pixel-based image classification which is grouped into a cover class for each pixel. This classification groups into 4 benthic habitat classes, namely dead coral sand (KMP), sea grass (LM), live coral sand (PKH) and sand (PS). The results showed that the application of the lyzenga correction using the MLH algorithm in mapping benthic habitats resulted in a significant accuracy rate (80,00%), while the application without lyzenga correction using the MLH algorithm resulted in an accuracy rate (76,00%). These findings support the use of the lyzenga correction method which can increase the level of accuracy by (4,00%) with the MLH algorithm as a guide in better mapping of benthic habitat cover in Pengudang Village.

Glacial lakes of Sikkim Himalaya: their dynamics, trends, and likely fate—a timeseries analysis through cloud-based geocomputing, and machine learning

In the background of ongoing climate change, it is important to monitor the spatial and temporal changes of glacial lakes (GLs) since they influence snowmelt runoff, stream discharge, water resources, and glacial lake outburst flood (GLOF). However, accurate identification and mapping of GLs in the background of snow-clad mountains through visual interpretation of satellite data is a tedious and challenging assignment when multiyear time-series analysis is considered. To overcome this challenge, automated extraction of GLs in satellite images has been carried out in this study with the help of machine learning (ML). The novelty of this study is identification and tracking of GLs over three decades using ML and geospatial analysis using pixel-based image classification. For this, Random Forest Classifier (RFC) and Artificial Neural Network (ANN) were employed. The methodology is demonstrated here for the identification and mapping of GLs in the Sikkim Himalaya from 1987 to 2020 and for forecasting the possible fate of these GLs through time-series modelling. The geospatial time-series analysis using Google Earth Engine, ML classifiers, and GIS framework, has captured the dynamics of GLs in Sikkim and has revealed the spatial and temporal patterns in GLs’ dimensions as well as GLOF risk.

Multi-Scale Hybrid Spectral Network for Feature Learning and Hyperspectral Image Classification

Hyperspectral image (HSI) classification is an important concern in remote sensing, but it is complex since few numbers of labelled training samples and the high-dimensional space with many spectral bands. Hence, it is essential to develop a more efficient neural network architecture to improve performance in the HSI classification task. Deep learning models are contemporary techniques for pixel-based hyperspectral image (HSI) classification. Deep feature extraction from both spatial and spectral channels has led to high classification accuracy. Meanwhile, the effectiveness of these spatial-spectral methods relies on the spatial dimension of every patch, and there is no feasible method to determine the best spatial dimension to take into consideration. It makes better sense to retrieve spatial properties through examination at different neighborhood scales in spatial dimensions. In this context, this paper presents a multi-scale hybrid spectral convolutional neural network (MS-HybSN) model that uses three distinct multi-scale spectral-spatial patches to pull out properties in spectral and spatial domains. The presented deep learning framework uses three patches of different sizes in spatial dimension to find these possible features. The process of Hybrid convolution operation (3D-2D) is done on each selected patch and is repeated throughout the image. To assess the effectiveness of the presented model, three benchmark datasets that are openly accessible (Pavia University, Indian Pines, and Salinas) and new Indian datasets (Ahmedabad-1 and Ahmedabad-2) are being used in experimental studies. Empirically, it has been demonstrated that the presented model succeeds over the remaining state-of-the-art approaches in terms of classification performance.

Mapping global distribution of mangrove forests at 10-m resolution

Mangrove forests deliver incredible ecosystem goods and services and are enormously relevant to sustainable living. An accurate assessment of the global status of mangrove forests warrants the necessity of datasets with sufficient information on spatial distributions and patch patterns. However, existing datasets were mostly derived from ∼30 m resolution satellite imagery and used pixel-based image classification methods, which lacked spatial details and reasonable geo-information. Here, based on Sentinel-2 imagery, we created a global mangrove forest dataset at 10-m resolution, namely, High-resolution Global Mangrove Forests (HGMF_2020), using object-based image analysis and random forest classification. We then analyzed the status of global mangrove forests from the perspectives of conservation, threats, and resistance to ocean disasters. We concluded the following: (1) globally, there were 145,068 km2 mangrove forests in 2020, among which Asia contained the largest coverage (39.2%); at the country level, Indonesia had the largest amount of mangrove forests, followed by Brazil and Australia. (2) Mangrove forests in South Asia were estimated to be in the better status due to the higher proportion of conservation and larger individual patch size; in contrast, mangrove forests in East and Southeast Asia were facing intensive threats. (3) Nearly, 99% of mangrove forest areas had a patch width greater than 100 m, suggesting that nearly all mangrove forests were efficient in reducing coastal wave energy and impacts. This study reports an innovative and up-to-date dataset and comprehensive information on mangrove forests status to contribute to related research and policy implementation, especially for supporting sustainable development.

Mapping Sea Grass Coverage of Tanjung Benoa Bali Using Medium Resolution Satellite Imagery Sentinel 2B

Seagrass beds are important components of a coastal ecosystem. This ecosystem serves as the primer producers of the water food chain, habitat for marine biota, produces organic carbon, and indirectly contributes to the economic well-being of coastal communities. However, the ecosystem is vulnerable to damage caused by natural factors and human activities. The objectives of this study were, firstly to identify the distribution of seagrass beds in Tanjung Benoa using Sentinel 2B satellite imagery and secondly to compare classification results from two different approaches namely pixel-based image classification and object-based image classification. Accuracy-test was carried out using field data reference of 195 sample points in the form of a 10 m X 10 m transect. The image pre-processing process was conducted with Bottom of Atmosphere (BoA) correction using the Dark Object Subtraction (DOS) method. Furthermore, the water column correction was performed using the Depth Invariant Index (DII) and the Lyzenga algorithm. The mapping results showed that the area of seagrass beds in the shallow waters of Tanjung Benoa reaches 242.99 ha. There were seven seagrass species in the study area, with an average cover of 75%. The accuracy of object-based image classification was higher than that of pixel-based classification with a difference up to 25% for six classes classification and 15% for two classes classification. Excellent results for classifying seagrasses based on cover density can be obtained when high-resolution satellite imagery and OBIA are combined with the SVM or Fuzzy Logic algorithm.

Spatial prediction of the urban inter-annual land surface temperature variability: An integrated modeling approach in a rapidly urbanizing semi-arid region

The inter-annual land surface temperature (LST) is a meteorological indicator of urban environments, affecting energy consumption and quality of life. In this study, the annual LST variability (ALSTV) of a rapidly urbanizing region in Iran including Karaj, Shahriar and Mohammad-Shahr cities and their surrounding suburbs was estimated using Landsat-8 images. Pixel-based image classification and object-based segmentation techniques were employed to extract built-up patches and spectrally homogeneous regions. Using the multiple regression analysis, the ALSTV of built-up patches was modeled as a function of their structure and the spatial characteristics of other land-use patches located in their respective segments (as neighborhood factors). A Cellular Automata model was utilized to simulate the expansion of built-up area up to 5% and, therefore, estimate their future ALSTV. The mean ALSTV was highest in urban areas (33.25± 7.41 °C), while the lowest mean value of 24.12± 3.24 °C was in green covers. The ALSTV of built-up patches were positively associated with their area and the percentage and size of neighborhood built-up patches, while negatively with the percentage of neighborhood green patches. To create a thermally comfortable landscape, additional green patch allocation and preventing excessive urban patch growth are required in all large built-up segments.

SO SÁNH CÁC PHƯƠNG PHÁP PHÂN LOẠI ẢNH VỆ TINH LANDSAT-8 ĐỂ XÂY DỰNG BẢN ĐỒ LỚP PHỦ MẶT ĐẤT TẠI THÀNH PHỐ HẠ LONG, TỈNH QUẢNG NINH

For a long time, many pixel-based image classification algorithms have been developed for identifying land cover, some of which are commonly used due to their efficiency and accuracy such as Maximum Likelihood (MLC), Support Vector Machines (SVMs), and Decision Trees (DTs). These methods are applied to classify land cover in Ha Long city using Landsat-8 satellite images with several categories including residence, bare soil, forest, agricultural land, water surface, and coal field. The validation results show that the overall accuracy (OA) and Kappa coefficient (K) of these classification methods are high, with OA > 91 % and K > 0.9. However, compared to the other two methods, the DTs method provides the results with the highest accuracy and the best ability to separate features. The obtained results allow selecting the image classification method for identical areas with complicated land cover such as Ha Long.

Assessment of Land Use-land Cover Change in Irga River Catchment Using Object-based Image Classification Technique

Land use-land cover (LULC) change analysis is essential for understanding the spatial and temporal change of landscape during a known long period for sustainable management of natural resources. The main objective of this study was to assess land use-land cover change using an object-based image classification technique which is a recent image classification technique with better accuracy than traditional pixel based image classification. The study was conducted in the catchment area of the Irga River, a tributary of the Barakar River, which falls in the Giridih district of Jharkhand (India). The catchment of the study area was delineated using SRTM DEM data (30 m spatial resolution). LANDSAT images (TM and OLI-TIRS) were used to develop the land use- land cover maps of 1997, 2007, and 2017 using object-based image analysis (OBIA). The images were classified and analyzed using ArcGIS and eCognition Developer 64 software. The accuracy of the classified images for each year was assessed by preparing the error matrix and calculating the Kappa coefficient. The overall accuracies of classified images were computed to be 88%, 83% and 91% while Kappa coefficients were found to be 0.8455, 0.7706 and 0.8796 for year 1997, 2007 and 2017 respectively. Over the 20 years (1997-2017), agricultural land increased by 12.23%, settlement increased by 76.62%, wasteland decreased by 39.59%, vegetation increased by 14.83%, water-bodies increased by 26.29%, and river area decreased by 16.66%. The analysis indicated an increasing trend in agricultural land, settlement, and vegetation while decreasing trends in wasteland and river areas. However, no definite trend was observed in the extent of the water-bodies.The results indicated that waste land greatly reduced and converted into settlement and agricultural land in the catchment.

Object-based characterization of vegetation heterogeneity with sentinel images proves efficient in a highly human-influenced National Park of Côte d’Ivoire

Forest monitoring requires more automated systems to analyze high ecosystem heterogeneity. The traditional pixel-based detection method has proven to be less and less effective. A novel change detection method is therefore proposed to detect changes in forest cover using satellite images at very high spatial resolution. This is object-oriented classification, which groups pixels into interpreted objects, based on their spectral values, spatial, and textural properties. Using sentinel and Lansat images, we tested for the first time in the West African rainforest zone the effectiveness of this method for better detection, delineation, and analysis of land use and occupation types. The mean shift algorithm was used in both the segmentation and classification processes. Next, we compared the proposed object-oriented method with a pixel-based image classification detection method by implementing both methods under the same conditions. High detection accuracy (> 90%) and an overall Kappa greater than 0.90 were obtained by the object-oriented method, which is about 20% higher than the pixel-based method. The object-based method was free of salt and pepper effects and was less prone to image misregistration in terms of change detection accuracy and mapping results. This study demonstrates that the object-based classifier is a much better approach than the classical pixel-based classifier. In addition, it shows the problems of detecting heterogeneous landscapes and explains the observed confusions between the types of vegetation formations specific to tropical wetlands. The results obtained are encouraging and the contribution of high-resolution images and the object-based method to better discrimination of tropical wetland vegetation is discussed.

Burned Area Detection Using Multi-Sensor SAR, Optical, and Thermal Data in Mediterranean Pine Forest

Burned area (BA) mapping of a forest after a fire is required for its management and the determination of the impacts on ecosystems. Different remote sensing sensors and their combinations have been used due to their individual limitations for accurate BA mapping. This study analyzes the contribution of different features derived from optical, thermal, and Synthetic Aperture Radar (SAR) images to extract BA information from the Turkish red pine (Pinus brutia Ten.) forest in a Mediterranean ecosystem. In addition to reflectance values of the optical images, Normalized Burn Ratio (NBR) and Land Surface Temperature (LST) data are produced from both Sentinel-2 and Landsat-8 data. The backscatter of C-band Sentinel-1 and L-band ALOS-2 SAR images and the coherence feature derived from the Interferometric SAR technique were also used. The pixel-based random forest image classification method is applied to classify the BA detection in 24 scenarios created using these features. The results show that the L-band data provided a better contribution than C-band data and the combination of features created from Landsat LST, NBR, and coherence of L-band ALOS-2 achieved the highest accuracy, with an overall accuracy of 96% and a Kappa coefficient of 92.62%.

Deep learning approach for land use images classification

CNN (convolutional neural networks) are a category of neural networks that are majorly used for image classification and recognition. This Deep Learning (DL) technique is used to solve complex problems, particularly for environmental protection, its approaches have affected several domains without exception, geospatial world is one vised domain. In this paper we aim to classify aerial images of Tangier region, city located in north of Morocco, by using pixel based image classification with convolutional Neural Networks. Flickr API is used to get our test images dataset. These images are used as input to a pretrained network Resnet18, a small convolution neural network architecture, which is able to recognize 21 land use classes of images. Our methodology is based on the following steps, first we set up the data, and then we re-train the cited Deep Learning model (Transfer Learning) and perform a quick and visual verification, by generating a labeled map from the geotagged images, labels correspond to class provided by the CNN neural network.

Attention-Based Multiscale Sequential Network for PolSAR Image Classification

Polarimetric synthetic aperture radar (PolSAR) images classification is an important topic for PolSAR images understanding and interpretation. However, traditional pixel-based PolSAR image classification that takes image pixel as a processing unit cannot make full use of spatial information and, thus, may not obtain the satisfactory classification results. Hence, this letter proposed an attention-based multiscale sequential network for PolSAR images classification increasing the multiscale spatial information between pixels by way of spatial sequence. Specifically, the long short-term memory (LSTM) network is introduced to convert the time sequence into spatial sequence to extract the spatial features. Then, to obtain the more abundant spatial features and select more important spatial information, an attention-based multiscale spatial-enhanced LSTM (AMSE-LSTM) is proposed to enhance the relationship between pixel spatial information. Finally, a new mixed loss function is defined to improve the classification performance. Experimental results with two real PolSAR data show that compared with state-of-the-art methods, the proposed method can achieve a much better performance and overall classification accuracy.

Kernel-Based Constrained Energy Minimization for Hyperspectral Mixed Pixel Classification

One fundamental task of hyperspectral imaging is spectral unmixing. In this case, the conventional pure pixel-based hyperspectral image classification (HSIC) may not work effectively for mixed pixels. This article presents a kernel-based approach to hyperspectral mixed pixel classification (HMPC) which includes two nonlinear mixed pixel classifiers, kernel constrained energy minimization (KCEM) and kernel linearly constrained minimum variance (KLCMV) to replace the widely used pure pixel-based support vector machine (SVM) classifier. Interestingly, what the binary-class and multiclass SVM classifiers are to pure pixel-based HSIC can be similarly derived for what a single-class KCEM detector and a multiclass KLCMV detector are to HMPC. In particular, the commonly used discrete classification map-based hard classification measures, average accuracy (AA) and overall accuracy (OA) for performance evaluation can be further generalized to real-valued mixed class abundance fractional map-based soft classification measures via 3-D receiver operating characteristic (3-D ROC) analysis-derived detection measures. Extensive experiments are conducted to demonstrate the utility of HMPC where KCEM/KLCMV not only significantly improve the classification performance of CEM/LCMV-based classifiers but also outperform many existing spectral-spatial classification methods.

GIS-BASED RAPID EARTHQUAKE EXPOSURE AND VULNERABILITY MAPPING USING LIDAR DEM AND MACHINE LEARNING ALGORITHMS: CASE OF PORAC, PAMPANGA

Abstract. Disaster risk reduction and management (DRRM) not only requires a thorough understanding of hazards but also knowledge of how much built-up structures are exposed and vulnerable to a specific hazard. This study proposed a rapid earthquake exposure and vulnerability mapping methodology using the municipality of Porac, Pampanaga as a case study. To address the challenges and limitations of data access and availability in DRRM operations, this study utilized Light Detection and Ranging (LiDAR) data and machine learning (ML) algorithms to produce an exposure database and conduct vulnerability estimation in the study area. Buildings were delineated through image thresholding and classification of the normalized Digital Surface Model (nDSM) and an exposure database containing building attributes was created using Geographic Information System (GIS). ML algorithms such as Support Vector Machine (SVM), logistic regression, and Random Forest (RF) were then used to predict the model building type (MBT) of delineated buildings to estimate seismic vulnerability. Results showed that the SVM model yielded the lowest accuracy (53%) while logistic regression and RF models performed fairly (72% and 78% respectively) as indicated by their F-1 scores. To improve the accuracy of the exposure database and vulnerability estimation, this study recommends that the proposed building delineation process be further refined by experimenting with more appropriate thresholds or by conducting point cloud classification instead of pixel-based image classification. Moreover, ground truth MBT samples should be used as training data for MBT prediction. For future work, the methodology proposed in this study can be implemented when conducting earthquake damage assessments.

Pixel-based vs. object-based anthropogenic impervious surface detection: driver for urban-rural thermal disparity in Faridabad, Haryana, India

Anthropogenic impervious surface area (AISA) expansion with its impact on temperature is current research trend. Majority of earlier research focused on land surface temperature (LST) versus AISA relationship in specific year independently. Therefore, present research is a novel attempt to study ΔLST–ΔAISA correlation during 2012–2017 in core urban, semi urban, and rural region of Faridabad. Results demonstrate applicability of object-based image classification (OBIC) over pixel-based image classification (PBIC) to detect AISA with higher accuracy (OBIC: 2012—94%, 2017—93%, PBIC: 2012—79%, 2017—84%). Degree of non-linearity of LST-AISA relationship increases from urban to rural region, evident from the coefficient of AISA% (0.03-core urban, 0.07-semi-urban, 0.08-rural). Absolute temperature is low in land with shallow ground water depth up to 10 m. Higher percentage of land with shallow water depth causes significant increase in (ΔLST/ΔAISA) in rural region. Consequently, authors suggest avoiding acquisition of land with water table depth 3–10 m for future impervious surface development.

Using Pixel-Based Image Classification to Identify City Structure Development in Balikpapan City, East Kalimantan

A city structure is a composition of forming a city. The shape of its city can characterize the development. City development must be sustainable, then the city structure formed must also support sustainable development. Pixel-based image classification is a method used in remote sensing to quickly and multi-temporally identify structures of development of the city. Balikpapan is a National Activity Center (PKN) that has a strategic location for national and international transport routes and, therefore, can accelerate the development of the city. The study aims to identify the development of city structures formed in Balikpapan from 2009, 2014, and 2019 by using pixel-based image classification and overlay analysis. The variables of this study are developed land, non-developed land, vegetation land, and water body. The study results show the development of the city structure by forming several core structures in Balikpapan, or called multi-nuclei, on city structure theory by Harris and Ullman in 1945. The built-up area dominates the development of the city structure, and development dominates from southern to eastern of Balikpapan City. The city center formed has a role as Central Business District (CBD), and the sub-city center created helps the city center to equalize sustainable development for a better future. The development of the city structure in Balikpapan provides direction to help adjust the city in the future.

Image translation between high-resolution optical and synthetic aperture radar (SAR) data

ABSTRACT This paper presents a novel study: remote-sensing image translation between high-resolution optical and Synthetic Aperture Radar (SAR) data through machine learning approaches. To this end, conditional Generative Adversarial Networks (cGANs) have been proposed with the guide of high-level image features. Efficiency of the proposed methods have been verified with different SAR parameters on three regions from the world: Toronto, Vancouver in Canada and Shanghai in China. The generated SAR and optical images have been evaluated by pixel-based image classification with detailed land cover types including: low and high-density residential area, industry area, construction site, golf course, water, forest, pasture and crops. Results showed that the translated image could effectively keep many land cover types with compatible classification accuracy to the ground truth data. In comparison with state-of-the-art image translation approaches, the proposed methods could improve the translation results under the criteria of common similarity indicators. This is one of first study on multi-source remote-sensing data translation by machine learning.

Land Use Changes Due to Mining Activities in Penajam Paser Utara Regency, East Kalimantan Province

There are coal mines in East Kalimantan Province, one of which is in Penajam Paser Utara Regency. The existence of a coal mine, resulting in changes in land use around it. If there is a change in land use that is not following the land potential, it will have an impact on the surrounding environment. The purpose of this study is to identify changes in land use in 2009 - 2019 and analyzethe relationship between the mining area and the area of change in land use. The data required is in the form of land use in 2009 and 2019, obtained based on interpretation using Landsat imagery and verification of existing land uses. The images used are Landsat 5 TM and SPOT 7 MS ORT images. To identify land use changes, the data processing method used is pixel based image classification with the supervised maximum likelihood classification technique. The analysis method used is an overlay. The result of this research is a decrease in the area of water body, forest, non built area, and farm. There is an increase in the area of built-up area and mining area. There is a relationship between mine area and changes in land use.

WITHDRAWN: Modified color pixel based image segmentation using fbmc algorithms used with real time image database

Abstract Image mining is an astonishing concept for data mining. For image mining, prior information is more essential to understand the data mining concept. Image mining is associated with extracting implicit information, associations with image data or other patterns not embedded directly in the images. It is the method of analyzing large collections of domain-specific data and consequently extracting information and insights for the decision-making process in the form of new relationships, patterns or clusters. Tiger become a reserved animal. Conservation of tiger has been a challenging task. This work would add a small account to the herculean task of conserving the species. Several scientific researchers have carried out their research on the tiger reserve conservation. This research work proposes a method to find the age of the tiger, using color as a parameter. Color pixel based image classification and clustering techniques has been used to identify the age of the tiger. This research work mainly focuses on RGB color spaces, which is implemented on the real time tiger images. The objective of the research work is to be done on assessing the age of the tiger using the color pixel based image classification and clustering is the main task of the research work and to optimize the image filtering and enhancement methods that are used to remove the noise and to improve the quality of pixels or images and assessing the processing Time, Retrieval Time, Accuracy and Error Rate by generating the better results is real time tiger image database.

UAV-Based RGB Imagery for Hokkaido Pumpkin (Cucurbita max.) Detection and Yield Estimation

Pumpkins are economically and nutritionally valuable vegetables with increasing popularity and acreage across Europe. Successful commercialization, however, require detailed pre-harvest information about number and weight of the fruits. To get a non-destructive and cost-effective yield estimation, we developed an image processing methodology for high-resolution RGB data from Unmanned aerial vehicle (UAV) and applied this on a Hokkaido pumpkin farmer’s field in North-western Germany. The methodology was implemented in the programming language Python and comprised several steps, including image pre-processing, pixel-based image classification, classification post-processing for single fruit detection, and fruit size and weight quantification. To derive the weight from two-dimensional imagery, we calculated elliptical spheroids from lengths of diameters and heights. The performance of this processes was evaluated by comparison with manually harvested ground-truth samples and cross-checked for misclassification from randomly selected test objects. Errors in classification and fruit geometry could be successfully reduced based on the described processing steps. Additionally, different lighting conditions, as well as shadows, in the image data could be compensated by the proposed methodology. The results revealed a satisfactory detection of 95% (error rate of 5%) from the field sample, as well as a reliable volume and weight estimation with Pearson’s correlation coefficients of 0.83 and 0.84, respectively, from the described ellipsoid approach. The yield was estimated with 1.51 kg m−2 corresponding to an average individual fruit weight of 1100 g and an average number of 1.37 pumpkins per m2. Moreover, spatial distribution of aggregated fruit densities and weights were calculated to assess in-field optimization potential for agronomic management as demonstrated between a shaded edge compared to the rest of the field. The proposed approach provides the Hokkaido producer useful information for more targeted pre-harvest marketing strategies, since most food retailers request homogeneous lots within prescribed size or weight classes.