Numerous uses of the hyperspectral remote sensing technology exist for identifying land cover and tracking its evolution. The classification of hyperspectral images must now take into account both spectral and spatial...
详细信息
Numerous uses of the hyperspectral remote sensing technology exist for identifying land cover and tracking its evolution. The classification of hyperspectral images must now take into account both spectral and spatial information due to recent advancements and the production of images with high spatial resolution. Convolutional neural networks (CNNs) have much employed in recent years to enhance the classification precision of hyperspectral images. The simultaneous use of spatial feature extraction methods in CNNs has not received significant attention in prior studies. In this study, a novel CNN architecture has been developed for classifying hyperspectral images. The weighted genetic (WG) algorithm is used in the proposed technique to minimize the hyperspectral image's dimensions. The WG algorithm keeps every band in the image and gives each one weight between zero and one based on how much information it contains. Following the expectation maximization (EM) method to the collected features, the segmented objects are then categorized using the CNN algorithm. Three benchmark hyperspectral images, Pavia, DC Mall, and Indiana Pine, were used to assess the proposed approach. The trials' findings demonstrate the proposed approach's superiority over the multilayer perceptron (MLP) algorithm in the Pavia, DC Mall, and Indiana Pine images by 14, 16, and 8% in the overall accuracy parameter, respectively.
A geological map is an indispensable instrument in the field of geological research. In the past, the sole method employed for the production of geological maps was manual mapping. This approach is costly in terms of ...
详细信息
A geological map is an indispensable instrument in the field of geological research. In the past, the sole method employed for the production of geological maps was manual mapping. This approach is costly in terms of time, resources and the amount of work required. Since the conclusion of the 20th century, remote sensing techniques, which entail the classification of satellite images, have been employed for the purpose of geological mapping to complement field work. The pixel method, which classifies based on pixels, is the most prevalent approach for the production of geological maps due to its greater ease of implementation. In contrast, object-based classification methods which consider sets of pixels (objects), are seldom employed, despite their considerable potential for geological mapping. The aim of this article is to compare the performance of these two types of satellite image classification for the update of geological maps (pixel and object-based method). The Central Highlands of Madagascar region has diverse geological, volcanic, sedimentary and metamorphism formations. The richness of geological features make harder geologic mapping. The use of Sentinel-1 and 2 images was justified by their superior accuracy and a high spatial resolution, permitting an easier details highlighting. The object-based method yielded geological map of notable accuracy, a slight superior Kappa coefficient (0.88 vs. 0.83). The generated map exhibit greater homogeneity and are more readily interpretable. Enabling, more straightforward identification of the different lithologies and easier geodynamic history comprehension of the study area.
作者:
Wang, WenyeZhang, XueliangXiao, PengfengSu, QiNanjing Univ
Sch Geog & Ocean Sci Jiangsu Prov Key Lab Geog Informat Sci & Technol Key Lab Land Satellite Remote Sensing Applicat Min Nanjing Jiangsu Peoples R China Nanjing Univ
Sch Geog & Ocean Sci Jiangsu Prov Key Lab Geog Informat Sci & Technol Key Lab Land Satellite Remote Sensing Applicat Min Nanjing 210023 Jiangsu Peoples R China
The application of deep learning (DL) improves the accuracy of object-based classification in urban area, but the huge numbers of labelled samples required for training DL models are difficult to obtain. To address th...
详细信息
The application of deep learning (DL) improves the accuracy of object-based classification in urban area, but the huge numbers of labelled samples required for training DL models are difficult to obtain. To address that, we propose a novel semi-supervised object-based classification method that combines pseudo-labelling method and consistency regularization method, named as perturbed peer model (PPM). To evaluate the quality of unlabelled object samples, a new object-level joint confidence is designed to assess the confidence of the samples' prediction, which provides guidance for selecting unlabelled object samples in semi-supervised object-based classification. Instead of only using the high-confidence samples as usual, both the high-confidence samples with discriminative power and the low-confidence samples with a large range of structural features are exploited using pseudo-labelling method and consistency regularization method, respectively, facilitating the fusion of the two methods in the proposed PPM. In addition, two types of perturbations, dropout and difference augmentation, are integrated into the model to drive the consistency regularization method as well as to enhance the difference to facilitate the pseudo-labelling method. Experimental results show that the classification accuracy of PPM is better than the widely-used self-training, co-training, noisy-student, unsupervised data augmentation for consistency training, and FreeMatch methods in urban area.
The study aims to improve the classification and mapping of snow cover over the Himalayan region, which is essential for assessing water availability and understanding hydrological and climatic interactions. The norma...
详细信息
The study aims to improve the classification and mapping of snow cover over the Himalayan region, which is essential for assessing water availability and understanding hydrological and climatic interactions. The normalized difference snow index (NDSI) is a traditional digital classification method for snow cover mapping. However, it is not always effective in differentiating snow from other features such as water bodies and shadows of mountain hills. In this study, an improved methodology for snow cover mapping was developed using an object-based classification with NDSI and normalized difference water index (NDWI) over segmented objects instead of pixels to separate snow and water with reduced noise. Shepherd segmentation was used to generate spatially homogeneous objects associated with ground cover features. The study focused on the Chandra basin in Himachal Pradesh, India, using an Indian Remote Sensing Satellite (IRS-P6) LISS-III optical image from 30-09-2016. The developed framework was tested using an object-based NDSI classification and further improved with an object-based NDSI-NDWI classification and validated against a manually digitized snow cover map. Validation showed that the object-based NDSI-NDWI classification provided a significant improvement in snow cover mapping compared to traditional NDSI classification, reducing the overestimation of snow-covered areas by up to 6.14%. The developed methodology was executed in the Python environment with efficient computing power. This study demonstrates that an integrated analysis of object-based classification with NDSI and NDWI, can significantly improve snow cover mapping by separating non-snow features. The results of this study show that they have the potential to be extended to larger regions with snow cover.
object-based features like spectral, topographic, and textural are supportive to determine debris-covered glacier classes. The original feature space includes relevant and irrelevant features. The inclusion of all the...
详细信息
object-based features like spectral, topographic, and textural are supportive to determine debris-covered glacier classes. The original feature space includes relevant and irrelevant features. The inclusion of all these features increases the complexity and renders the classifier's performance. Therefore, feature space optimization is requisite for the classification process. Previous studies have shown a rigorous exercise in manually selecting the best combination of features to define the target class and proven to be a timeconsuming task. The present study proposed a hybrid feature selection technique to automate the selection of the best suitable features. This study aimed to reduce the classifier's complexity and enhance the performance of the classification model. Relief-F and Pearson Correlation filter-based feature selection methods ranked features according to the relevance and filtered out irrelevant or less important features based on the defined condition. Later, the hybrid model selected the common features to get an optimal feature set. The proposed hybrid model was tested on Landsat 8 images of debris-covered glaciers in Central Karakoram Range and validated with present glacier inventories. The results showed that the classification accuracy of the proposed hybrid feature selection model with a Decision Tree classifier is 99.82%, which is better than the classification results obtained using other mapping techniques. In addition, the hybrid feature selection technique has sped up the process of classification by reducing the number of features by 77% without compromising the classification accuracy. (c) 2022 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams University This is an open access article under the CC BY-NC-ND license (http://***/licenses/by-ncnd/4.0/).
A semi-automated method has been developed for the extraction of land degradation processes using multi sensor data by applying an object-based classification. The object-based approach creates homogenous objects, whi...
详细信息
A semi-automated method has been developed for the extraction of land degradation processes using multi sensor data by applying an object-based classification. The object-based approach creates homogenous objects, which is the key component of this classification. The study utilized optical satellite (Landsat-8), microwave (RISAT-1, SAR) and Cartosat-1 digital elevation model (DEM) over Kanpur Dehat district, Uttar Pradesh, and Surendranagar district, Gujarat, India. The objects were created using Shepherd segmentation algorithm. Normalized difference vegetation index (NDVI) was used to classify the degraded and no apparent degradation (NAD) objects based on the three seasons (rabi, summer, and kharif) Landsat-8 bands. Degraded objects were further classified into salinity, forest water erosion, and water logging using brightness index based on Landsat-8, proximity analysis near the river channel using RISAT-1, and low-lying area using DEM, respectively. The digitally generated results were validated with manual digitized desertification status maps (DSM) published by Space Applications Centre, Ahmedabad, India. The overall accuracy and kappa coefficient for Kanpur Dehat and Surendranagar districts were found 84.67%, 0.79 and 72.33%, 0.60, respectively. This study was carried out based on integrated analysis of different satellites (optical, microwave, and DEM). The advantage of newly designed framework offers less chance of mixing and narrowing down of the area for further classification with better accuracy. The developed framework is based on analytical approach, which was tested and implemented in the Python environment with efficient computing power. The study illustrates that the developed approach is independent of climatic-topographic conditions and executed over pilot study sites, which could be extended over larger regions of the land use/land cover for land degradation mapping.
Urban building information can be effectively extracted by applying object-based image segmentation and multi-stage thresholding on High Resolution (HR) remote sensing satellite imageries. This study provides the resu...
详细信息
Urban building information can be effectively extracted by applying object-based image segmentation and multi-stage thresholding on High Resolution (HR) remote sensing satellite imageries. This study provides the results obtained using this method on the images of Indian remote sensing satellite, CARTOSAT-2S launched by the Indian Space Research Organization (ISRO). In this study, a method is developed to extract urban building footprints from the HR remote sensing satellite images. The first step of the process consists of generating highly dense per pixel Digital Surface Model (DSM) by using semi global matching algorithm on HR satellite stereo images and applying robust ground filtering to generate Digital Terrain Model (DTM). In the second step, multi-stage object-based approach is adopted to extract building bases using the PAN sharpened image, normalized Digital Surface Model (nDSM) derived from DSM and DTM, and Normalised Difference Vegetation Index (NDVI). The results are compared with the manual method of drawing building footprints by cartographers. An average precision of 0.930, recall of 0.917, and f-score of 0.922 are obtained. The results are found to be in a match with the method using the high resolution Airborne LiDAR DSM by providing a solution for large areas, low cost and low time.
This paper presents a new framework for object-based classification of high-resolution hyperspectral *** multi-step framework is based on multi-resolution segmentation(MRS)and Random Forest classifier(RFC)*** first st...
详细信息
This paper presents a new framework for object-based classification of high-resolution hyperspectral *** multi-step framework is based on multi-resolution segmentation(MRS)and Random Forest classifier(RFC)*** first step is to determine of weights of the input features while using the object-based approach with MRS to processing such *** the high number of input features,an automatic method is needed for estimation of this ***,we used the Variable Importance(VI),one of the outputs of the RFC,to determine the importance of each image ***,based on this parameter and other required parameters,the image is segmented into some homogenous ***,the RFC is carried out based on the characteristics of segments for converting them into meaningful *** proposed method,as well as,the conventional pixel-based RFC and Support Vector Machine(SVM)method was applied to three different hyperspectral data-sets with various spectral and spatial *** data were acquired by the HyMap,the Airborne Prism Experiment(APEX),and the Compact Airborne Spectrographic Imager(CASI)hyperspectral *** experimental results show that the proposed method is more consistent for land cover mapping in various *** overall classification accuracy(OA),obtained by the proposed method was 95.48,86.57,and 84.29%for the HyMap,the APEX,and the CASI datasets,***,this method showed better efficiency in comparison to the spectralbasedclassifications because the OAs of the proposed method was 5.67 and 3.75%higher than the conventional RFC and SVM classifiers,respectively.
This paper explores the potentiality of using the completed local binary pattern (CLBP) for the classification of an urbanized oasis area situated in southeastern Tunisia, in very high spatial resolution GeoEye imager...
详细信息
This paper explores the potentiality of using the completed local binary pattern (CLBP) for the classification of an urbanized oasis area situated in southeastern Tunisia, in very high spatial resolution GeoEye imagery. To further improve the spatial information description derived by CLBP, which is successfully used in face recognition, we applied the multi-scale completed local binary pattern (MSCLBP) to classify the single-dated image. A supervised classification framework preceded by mean-shift segmentation is applied using the texture features alone and combined with the normalized difference vegetation index (NDVI). As the segmentation is a crucial step to obtain a good mapping, considerations are given to select the optimal combination of mean-shift input parameters, such as spatial radius and range radius. The results of this study indicate that MSCLBP outperforms the single-scale CLBP and the gray-level co-occurrence matrix (GLCM) descriptors in the task of expressing the classes of interests.
Using satellite data to extract forest structure mapping parameters assists forest management. In this research, structural parameters including species, density, canopy, and gaps were extracted from SPOT-7 satellite ...
详细信息
Using satellite data to extract forest structure mapping parameters assists forest management. In this research, structural parameters including species, density, canopy, and gaps were extracted from SPOT-7 satellite data over Hyrcanian forests (Iran). A detailed ground inventory was initially conducted, over 12 x 1 ha (100 m x 100 m) plots, in which tree coordinates were plotted, using a differential global positioning system (DGPS), along with data on tree species, diameter-at-breast-height and height, as well as canopy dimensions, and canopy gap shapes, sizes, and positions, for each plot. Then, spectral transformations, vegetation indices, and simple spectral ratios were extracted from SPOT-7 data, and a supervised, pixel-basedclassification method and a support-vector machine algorithm were used to classify and determine tree species types. In addition, canopy tree borders and gaps were classified, using an object-based method, and tree densities per unit area were determined, using the canopy gravity center. Finally, the original ground data was used to perform an accuracy assessment on the extracted information, with the results showing that forest type could be determined with 95% accuracy and a Kappa coefficient of 0.8. Canopy and gap coverage achieved an overall accuracy of 91% (Kappa coefficient: 0.7), and tree densities per hectare were determined, on average, to be 47 trees fewer than reality. In conclusion, we have shown that forest structural parameters could be extracted, with good accuracy, using a combination of pixel- and object-based methods applied to SPOT-7 imaging.
暂无评论