Sub-pixel Classification Research Articles

Super resolution land cover mapping of hyperspectral images using the deep image prior-based approach

ABSTRACTDue to the instantaneous field-of-view (IFOV) of the sensor and diversity of land cover types, some pixels, usually named mixed pixels, contain more than one land cover type. Soft classification can predict the portion of each land cover type in mixed pixels in the absence of spatial distribution. The spatial distribution information in mixed pixels can be solved by super resolution mapping (SRM). Typically, SRM involves two steps: soft class value estimation, which is similar to the image super resolution of image restoration, and land cover allocation. A new SRM approach utilizes a deep image prior (DIP) strategy combined with a super resolution convolutional neural network (SRCNN) to estimate fine resolution fraction images for each land cover type; then, a simple and efficient classifier is used to allocate subpixel land cover types under the constraint of the generated fine fraction images. The proposed approach can use prior information of input images to update network parameters and no longer require training data. Experiments on three different cases demonstrate that the subpixel classification accuracy of the proposed DIP-based SRM approach is significantly better than the three conventional SRM approaches and a transfer learning-based neural network SRM approach. In addition, the DIP-SRM approach performs very robustly about small-area objects within multiple land cover types and significantly reduces soft classification uncertainty. The results of this paper provide an extension for utilizing SRCNN to address SRM issues in hyperspectral images.

Assessment of urban growth using Shannon’s Entropy Index: A case study of Chennai, Detroit of India

The Remote Sensing (R/ S) and Geographical Information System (GIS) play a vital role to evaluate and study the urban expansion pattern. In this study, the Chennai city was selected to perform the urban sprawl study. Five different periods of satellite imageries for the time elapsed between 1994 and 2016 were used. The main aim of this paper was to identify the urban sprawl of Chennai as a patterning process. The extended areas of urban in the period of 1994, 2001, 2006, 2011 and 2016 were extracted by the sub-pixel classification method from the satellite imageries. Furthermore, Shannon’s entropy index was used for assessing urban expansion. The findings of this study proved that Chennai city has sprawled by urban expansion during the period between 1994 and 2016. Likewise, the dispersion rate of urban sprawl for the periods of 1994, 2001, 2006, 2011 and 2016 were 0.671, 0.679, 0.688, 0.693 and 0.695 respectively. Consequently, this uncontrolled dispersed urban development had resulted in the study area losses their green space.

Sub-Pixel Crop Type Classification Using PROBA-V 100 m NDVI Time Series and Reference Data from Sentinel-2 Classifications

This paper presents the results of a sub-pixel classification of crop types in Bulgaria from PROBA-V 100 m normalized difference vegetation index (NDVI) time series. Two sub-pixel classification methods, artificial neural network (ANN) and support vector regression (SVR) were used where the output was a set of area fraction images (AFIs) at 100 m resolution with pixels containing estimated area fractions of each class. High-resolution maps of two test sites derived from Sentinel-2 classifications were used to obtain training data for the sub-pixel classifications. The estimated area fractions have a good correspondence with the true area fractions when aggregated to regions of 10 × 10 km2, especially when the SVR method was used. For the five dominant classes in the test sites the R2 obtained after the aggregation was 86% (winter cereals), 81% (sunflower), 92% (broad-leaved forest), 89% (maize), and 67% (grasslands) when the SVR method was used.

Effective Morphological Transformation and Sub-pixel Classification of Clustered Images

The main aim of this research work is to perform the morphological operations with reduced time complexity and area complexity. Morphological operation is the key element in any image processing. Finding the maximum and minimum using a window of defined size will imply to the morphological dilation and erosion respectively. So the proposed algorithm should be fast in the comparison and sorting, this way the time complexity could be reduced. It’s believed that the anchor concept will fetch this cause. The idea behind this is it fixes a pixel and setting it as the center pixel all the surrounding pixels will be processed. Moreover this is now been implemented for rectangular structuring element. This paper attempts the same for flat and 3D structuring elements. Hyper-spectral Imaging is a developing zone of remote detecting applications. Hyper-spectral pictures incorporate more extravagant and better otherworldly data than the multi-spectral pictures got previously. Hyper-otherworldly pictures are described by an exchange off between the unearthly and spatial resolution. The principle issue of the hyper-ghostly information is the generally low spatial goal. For arrangement, the serious issue brought about by low spatial goal is the blended pixels. Blended pixels alluded to the pixels which are involved by more than one land spread class. In the proposed procedure another strategy is utilized to address the issue of blended pixels and to get a better spatial goal of the land spread characterization maps. The strategy misuses the upsides of both picture bunching methods and phantom dimming calculations, so as to decide the fragmentary plenitudes of the classes at a sub-pixel scale. Spatial regularization by Flank planning method is at last performed to spatially find the got classes at sub-pixel level.

Long-Term Annual Surface Water Change in the Brazilian Amazon Biome: Potential Links with Deforestation, Infrastructure Development and Climate Change

The Brazilian Amazon is one of the areas on the planet with the fastest changes in forest cover due to deforestation associated with agricultural expansion and infrastructure development. These drivers of change, directly and indirectly, affect the water ecosystem. In this study, we present a long-term spatiotemporal analysis of surface water annual change and address potential connections with deforestation, infrastructure expansion and climate change in this region. To do that, we used the Landsat Data Archive (LDA), and Earth Engine cloud computing platform, to map and analyze annual water changes between 1985 and 2017. We detected and estimated the extent of surface water using a novel sub-pixel classifier based on spectral mixture analysis, followed by a post-classification segmentation approach to isolate and classify surface water in natural and anthropic water bodies. Furthermore, we combined these results with deforestation and infrastructure development maps of roads, hydroelectric dams to quantify surface water changes linked with them. Our results showed that deforestation dramatically disrupts small streams, new hydroelectric dams inundated landmass after 2010 and that there is an overall trend of reducing surface water in the Amazon Biome and watershed scales, suggesting a potential connection to more recent extreme droughts in the 2010s.

Unsupervised Nonlinear Spectral Unmixing of Satellite Images Using the Modified Bilinear Model

Episodes of mixing pixels in satellite imageries are more prevalent. Hence, spectral unmixing approach is used to perform the sub-pixel classification of satellite images. Many unmixing works were done based on the assumption that the pixels are linearly mixed (single interaction) but in real scenarios, the pixels are nonlinearly mixed due to interactions. Fan model and generalized bilinear model consider only the bilinear interactions for nonlinear unmixing. In reality, multiple interactions between the various classes are also present in the image. In this work, a new model, ‘modified bilinear model’ is proposed to perform the nonlinear unmixing process that considers the entire single, bilinear and multiple interactions into account. This system adaptively changes the mixing model on per pixel basis depending on the nonlinearity parameter. It has been tested with the multispectral, synthetic and real hyperspectral datasets and also illustrated notable advantages compared with the other methods.

Convolutional network architectures for super-resolution/sub-pixel mapping of drone-derived images

Spatial resolution enhancement is a pre-requisite for integrating unmanned aerial vehicle (UAV) datasets with the data from other sources. However, the mobility of UAV platforms, along with radiometric and atmospheric distortions, makes the task difficult. In this paper, various convolutional neural network (CNN) architectures are explored for resolving the issues related to sub-pixel classification and super-resolution of drone-derived datasets. The main contributions of this work are: 1) network-inversion based architectures for super-resolution and sub-pixel mapping of drone-derived images taking into account their spectral-spatial characteristics and the distortions prevalent in them 2) a feature-guided transformation for regularizing the inversion problem 3) loss functions for improving the spectral fidelity and inter-label compatibility of coarser to finer-scale mapping 4) use of multi-size kernel units for avoiding over-fitting. The proposed approach is the first of its kind in using neural network inversion for super-resolution and sub-pixel mapping. Experiments indicate that the proposed super-resolution approach gives better results in comparison with the sparse-code based approaches which generally result in corrupted dictionaries and sparse codes for multispectral aerial images. Also, the proposed use of neural network inversion, for projecting spatial affinities to sub-pixel maps, facilitates the consideration of coarser-scale texture and color information in modeling the finer-scale spatial-correlation. The simultaneous consideration of spectral bands, as proposed in this study, gives better super-resolution results when compared to the individual band enhancements. The proposed use of different data-augmentation strategies, for emulating the distortions, improves the generalization capability of the framework. Sensitivity of the proposed super-resolution and sub-pixel mapping frameworks with regard to the network parameters is thoroughly analyzed. The experiments over various standard datasets as well as those collected from known locations indicate that the proposed frameworks perform better when compared to the prominent published approaches.

Comparative performance of regression tree and parametric classification of savannah woody cover on SPOT 6 NAOMI imagery

Earth observation data potentially can provide ecosystem state indicative data for use in nature conservation. Full realisation of this potential requires robust algorithms that can extract the required ecosystem attributes from imagery. Woody cover is an indicative ecosystem state variable in savannah wilderness areas, whose extraction poses a challenge in reproducing the characteristically discontinuous cover. In this work the performance of the more modern, nonparametric regression tree and Random Forest classifiers in mapping savannah woody cover is compared with the parametric, potentially highly accurate sub-pixel classification approach. Detailed field quantification of % woody cover in 1 ha plots yielded data for the training stage of the two decision tree-based nonparametric classifiers. SPOT 6 NAOMI imagery whose timing excluded grass from the spectral signature of non-senescent vegetation was used. In order to improve the capability to extract woody cover the 6 m resolution of the multispectral bands was enhanced to 1.5 m, using high pass filter pan-sharpening. For the decision tree-based classifiers multiresolution image segmentation was then employed as preliminary step in extracting the woody cover. Objects approximating the 1 ha field plots were delineated at the coordinates of the respective field sampling sites. The univariate decision tree training stage used 60% of the sampling sites. The woody cover classes 0–19%, 20–39%, 40–59% and 60–79% were the labels of recursive partitioning terminal tree leaves. For Random Forest, up to 500 decision trees (Ntree) were specified. For regression tree modelling the 10-fold cross validation approach using the 1-SE rule was used for optimal tree selection, and pruning using 20% of the field sampling sites was accomplished using the cost complexity pruning technique. The remaining 20% of the field sampling sites were reserved for accuracy assessment, for which an additional stratified random sample of coordinates verified using a concurrent high resolution QuickBird image was used. Spectral signatures approximating Gaussian distributions were generated using woody end member sites for sub-pixel classification. Random Forest produced the highest overall accuracy (95.7%), followed by regression tree modelling (90.6%), and sub-pixel classification (73.3%). The two nonparametric classifiers correctly modelled the landscape distribution, occasional continuous formations and discontinuities in the woody cover. Based on the results, nonparametric classifiers appeared more appropriate for mapping the discontinuous savannah woody cover.

Automated Spectral Mapping and Subpixel Classification in the Part of Thar Desert Using EO-1 Satellite Hyperion Data

To evaluate the application potential of hyperspectral imaging, the automated spectral mapping and subpixel classification is used to classify the EO-1 satellite Hyperion data. Once the endmembers are extracted from the conventional hyperspectral processing steps (namely, atmospheric correction, noise reduction, pixel purity, and n-dimensional visualization), these endmembers are resolved for constituent rock minerals. Subsequently, the image is classified for mineralogy and lithology, and validated with the ground-truth data (prevailing geological map).

A novel bilinear unmixing approach for reconsideration of subpixel classification of land cover

Hyperspectral mixing depends on a variety of factors such as the instantaneous field of view of observation device, the height of the image capturing platform from the scene, the properties of materials in the pixel, and the interaction structure of these materials with incident light in the scene. Different unmixing models have been considered to model the hyperspectral mixing. The simplest one is the linear mixing model. Nevertheless, it has been recognized that mixing phenomena is usually nonlinear. Bilinear and linear–quadratic models have become popular recently, and also the bilinear polynomial post-nonlinear model shows promising results. Most of these nonlinear models consider only the reflection interaction. However, especially in regions like vegetated areas, absorbance and transmittance are also important facts which affect the mixture. So, they should be taken into account when dealing with physics of hyperspectral mixing. In this study, an enhanced bilinear mixing model is proposed for analyzing the physics of hyperspectral mixing. The model takes into account the transmittance, absorption and reflection. The results show that our enhanced bilinear viewpoint is superior, in terms of pixel reconstruction error, when compared to that of linear and other bilinear models which consider only the reflection interaction.

Tree canopy cover and its potential to reduce co2 in south of peninsular malaysia

Urban trees provide a wide range of ecosystem services that can address climate-change mitigation and adaptation. In this study, the tree cover and their potential to store carbon in two cities (Johor Bahru and Pasir Gudang) that are developing rapidly in the south of Peninsular Malaysia have been estimated. Tree coverage was mapped using Landsat 8 Thematic Mapper satellite data for year 2016. Various digital image processing techniques namely Maximum Likelihood and a sub-pixel classification were applied to obtain tree coverage of urban trees/forest, mangrove and oil palm. Results of the study show that natural tree coverage (forest and mangrove) in the cities range between 19 % and 47 % and generally Pasir Gudang has more tree coverage compared to Johor Bahru. Johor Bahru is the centre for various business and cultural activities, thus more built up areas are found in the city. On average, trees in the cities store approximately 796,136 t carbon or 2,919,164 t CO2-eq which is about 18 % of the total CO2–equivalent emissions projected for 2025 under the Business as Usual (BaU) scenario. The mapping of tree canopy cover and estimating their potential to store carbon is important for assessing climate change mitigation.

Integration of Contextual Knowledge in Unsupervised Subpixel Classification: Semivariogram and Pixel-Affinity Based Approaches

This letter investigates the use of coarse-image features for predicting class labels at a given finer spatial scale. In this regard, two unsupervised subpixel mapping approaches, a semivariogram method, and a pixel-affinity based method are proposed. Furthermore, segmentation-based spectral unmixing is explored so as to address the spectral variability and nonconvexity of classes. In addition, the gradient information is employed to resolve uncertainties in the unmixing process. The proposed modifications based on pixel-affinity and semivariogram have produced an accuracy improvement of 5% or more over the state-of-the-art approaches.

Forest Cover Monitoring and Change Detection in Nfifikh Forest (Morocco)

Peri-urban forests are subject to different dynamics due to several factors. Nfifikh forest is a man-made space, located in suburban of Mohammedia City, belonging to Casablanca, Settat Region, and geographically between Casablanca, the economic and business Capital of Morocco and Rabat, the national political capital. Over the past three decades, it has experienced several significant degradations. The aim of this study is to evaluate and quantify the deforestation within the study area using a forest cover change detection of various vegetation indices and subpixel classification to pick out high density plots with Landsat images TM, ETM+ and OLI. Remote sensing is used to highlight the changes caused through Space-Time. This monitoring might help managers to generate forest management plans and to moderate the speed of deforestation and degradation. The results show a significant change in vegetation cover detected between 1987 and 2015. The Density increased in 2001 while it decreased considerably in 2015.

A Subpixel Classification of Multispectral Satellite Imagery for Interpetation of Tundra-Taiga Ecotone Vegetation (Case Study on Tuliok River Valley, Khibiny, Russia)

The tundra–taiga ecotone plays significant role in northern ecosystems. Due to global climatic changes, the vegetation of the ecotone is the key object of many remote-sensing studies. The interpretation of vegetation and nonvegetation objects of the tundra–taiga ecotone on satellite imageries of a moderate resolution is complicated by the difficulty of extracting these objects from the spectral and spatial mixtures within a pixel. This article describes a method for the subpixel classification of Terra ASTER satellite image for vegetation mapping of the tundra–taiga ecotone in the Tuliok River, Khibiny Mountains, Russia. It was demonstrated that this method allows to determine the position of the boundaries of ecotone objects and their abundance on the basis of quantitative criteria, which provides a more accurate characteristic of ecotone vegetation when compared to the per-pixel approach of automatic imagery interpretation.

A Hybrid Subpixel Mapping Framework for Hyperspectral Images Using Collaborative Representation

Subpixel mapping with a low-resolution hyperspectral image as the only input is widely applicable due to the fact that auxiliary image is not always available in practice. In this paper, the collaborative representation-based subpixel mapping (CRSPM) framework is proposed to acquire an improved classification map at subpixel scale with only a low-resolution hyperspectral image available. To efficiently extract and utilize spatial information in this case without auxiliary image, the low-resolution hyperspectral (LHS) image is processed in a hybrid framework in two different ways to generate two subpixel scale classification maps. One is obtained by classifying the upsampled LHS image using collaborative representation-based (CR-based) classifier. The other is available using CR-based classification combined with spectral unmixing and subpixel spatial attraction model. Specifically, to enclose the contextual spatial information for higher classification accuracy, a spatially joint as well as post-partitioning CR-based classifier, JCRT-based classifier, is proposed and applied in this work. To achieve better classification performance, decision fusion is applied to determine class label from the two classification maps for each subpixel by the voting of the neighboring subpixels. Experimental results illustrate that the proposed CRSPM approach clearly outperforms some state-of-the-art subpixel mapping approaches by producing smoother classification map with less misclassification.

Sub-pixel classification of hydrothermal alteration zones using a kernel-based method and hyperspectral data; A case study of Sarcheshmeh Porphyry Copper Mine and surrounding area, Kerman, Iran

Remote sensing image analysis can be carried out at the per-pixel (hard) and sub-pixel (soft) scales. The former refers to the purity of image pixels, while the latter refers to the mixed spectra resulting from all objects composing of the image pixels. The spectral unmixing methods have been developed to decompose mixed spectra. Data-driven unmixing algorithms utilize the reference data called training samples and end-members. The performance of algorithms using training samples can be negatively affected by the curse of dimensionality. This problem is usually observed in the hyperspectral image classification, especially when a low number of training samples, compared to the large number of spectral bands of hyperspectral data, are available. An unmixing method that is not highly impressed by the curse of dimensionality is a promising option. Among all the methods used, Support Vector Machine (SVM) is a more robust algorithm used to overcome this problem. In this work, our aim is to evaluate the capability of a regression mode of SVM, namely Support Vector Regression (SVR), for the sub-pixel classification of alteration zones. As a case study, the Hyperion data for the Sarcheshmeh, Darrehzar, and Sereidun districts is used. The main classification steps rely on 20 field samples taken from the Darrehzar area divided into 12 and 8 samples for training and validation, respectively. The accuracy of the sub-pixel maps obtained demonstrate that SVR can be successfully applied in the curse of dimensional conditions, where the size of the training samples (12) is very low compared to the number of spectral bands (165).

The potential for integrating Sentinel 2 MSI with SPOT 5 HRG and Landsat 8 OLI imagery for monitoring semi-arid savannah woody cover

ABSTRACTMultitemporal archived imagery enables the monitoring of savannah woody cover, for ecological purposes. Compatibility in multitemporal, multiple sensor image data would facilitate the monitoring. The decommissioning of SPOT 5 (Système Pour l’Observation de la Terre 5) left a void in multispectral imagery at the 10 m spatial resolution of its high-resolution geometric (HRG) sensor. The subsequent launch of Sentinel 2 presented an opportunity for data continuity to monitor the savannah woody cover, using equivalent 10 m resolution multispectral instrument (MSI) bands. This study examined the integration potential of Sentinel 2 MSI with the longer archive HRG and Landsat 8 (Land Satellite 8) Operational Land Imager (OLI) imagery, in assessing savannah woody cover. Images of three semi-arid savannah sites acquired on same season dates that excluded herbaceous vegetation from the spectral signature were used: November 2014 (HRG) and December 2015 (MSI, OLI). Using equivalent green (G), red (R), and near infrared (NIR) bands at 10 m (MSI, HRG) and 30 m (OLI) resolution, the woody cover was mapped through subpixel classification. The mapped woody cover was compared for statistical differences using χ2 analysis at 10 m resolution (MSI, HRG) and at a degradation of the MSI and HRG images to the 30 m OLI pixel size. Conversion to top-of-atmosphere reflectance values facilitated inter-sensor correlation of G, R, and NIR reflectance for field sampling sites where woody cover was quantified. Inter-sensor regression functions in G, R, and NIR band MSI and HRG images were developed. The 10 m resolution classifications of woody cover were not statistically different. Due to spatial resolution similarity, SPOT 5 HRG multispectral imagery was established as suitable for integration with equivalent band MSI imagery in mapping the woody cover in a multitemporal analysis. For dense woody cover, Landsat 8 OLI imagery was more suitable for integration with MSI than HRG images due to higher radiometric sensitivity, which can permit monitoring physiology-related woody reflectance.

Estimation of Water-Spread Area of Singoor Reservoir, Southern India by Super Resolution Mapping of Multispectral Satellite Images

Estimation of reservoir water-spread area is often carried out by field surveys which are cumbersome, time consuming, expensive and involves more man power. Hence, such surveys cannot be carried out periodically. To overcome this issue, satellite images are used, wherein the reservoir water-spread is estimated by conventional per-pixel classification such as the maximum likelihood and minimum distance to mean approaches that often results in inaccurate estimate of water-spread area due to the presence of mixed pixels. High cost and non-availability of high resolution images demands the use of an alternative approach that can give accurate information about the reservoir water-spread area. In this work, IRS, LISS III images of moderate (24 m) resolution were used for accurate estimation of the water-spread area of Singoor reservoir, southern India. The reservoir water-spread areas were extracted using per-pixel classification, sub-pixel classification and super resolution mapping approaches. These results were validated with the water-spread areas obtained from field data of the same dates. The error produced by the per-pixel approach was 6.66%, while it was 4.37% for the sub-pixel approach and a meagre 1.71% for the super-resolution approach. Fairly less error produced by the super resolution mapping technique implies that it is an efficient approach for accurate quantification of reservoir water-spread area. The estimated water-spread can be used in a simple volume estimation formula to estimate the capacity of the reservoir.

A comparative study of signal transformation techniques in automated spectral unmixing of infrared spectra for remote sensing applications

ABSTRACTFrom geological and planetary exploration perspectives, automated sub-pixel classification of hyperspectral data is the most difficult task as it involves blind unmixing with library spectra of minerals. In this study, we demonstrate a procedure involving spectral transformation and linear unmixing to achieve the above task. For this purpose, infrared spectra of rocks from the spectral library, field, and remotely sensed hyperspectral image cube were used. Potential spectra of minerals for unmixing rock spectra were drawn from the library based on similarity of absorption features measured using Pearson correlation coefficient. Eight transformation techniques namely, first derivative, fast Fourier transform, discrete wavelet transform, Hilbert–Huang transform, crude low pass filter, S-transform, binary encoding, spectral effective peak matching, and two sparsity-based techniques (orthogonal matching pursuit, sparse unmixing via variable splitting, and augmented Lagrangian) were evaluated. Subsequently, minerals identified by above techniques were unmixed by linear mixture model (LMM) to decipher mineralogical composition and abundance. Results of LMM achieved using fully constrained least-square-estimation-based quadratic programming optimization approach were evaluated by conventional procedures such as X-ray diffraction and microscopy. In the case of image cube, endmembers derived using minimum noise fraction and pixel purity index were subjected to above procedure. It is evident that the discrete-wavelet-transformation-based approach produced excellent and meaningful results due to its flexibility in scaling the data and capability to handle noisy spectra. It is interesting to note that the adopted procedure could perform sub-pixel classification of image cube automatically and identify predominance of dolomite in limestone and sodium in alunite based on subtle differences in absorption positions.

Advanced Investigation of Remote Sensing to Geological Mapping of Zefreh Region in Central Iran

This study has tried to prove the ability of remote sensing techniques to extract information necessary for preparation of geological mapping of the earth’s surface using multi-spectral satellite images which are rich sources of Earth’s surface information. In this study, the surface geological mappings of Zefreh region have been investigated through ASTER, OLI, and IRS-PAN remote sensing data. To prepare the geological map, preprocessing steps and reducing noises from data using MNF algorithm were firstly carried out. Then a set of processing algorithms and image classification methods are included; the band rationing, color composite and pixel classification based on maximum likelihood, spectral and sub-pixel classification methods of spectral angle mapper (SAM), spectral feature fitting (SFF), linear spectral differentiation (LSU), hill-shade images and automatic lineament extraction were used. Confusion matrix was formed for all classified images through control points were randomly selected from 1:25,000 map of the region to determine the accuracy of obtained results, which indicated the maximum accuracy (up to 90%) of output images. Comparing the results obtained from these methods with the map prepared by ground operations confirmed accuracy results. Finally, the surface geology and fault map of Zafreh region was produced by combining detected geological formations and tectonic lineaments.