Final Classification Map Research Articles

Hierarchical classification for improving parcel-scale crop mapping using time-series Sentinel-1 data

Parcel-scale crop classification utilizing time-series satellite observations is of significant importance in precision agriculture. The prior knowledge that crop types can be organized in a hierarchical tree structure is beneficial for improving crop classification. Moreover, the crop hierarchy aligns with the coarse-to-fine cognitive process of geographic scenes. Based on the crop hierarchy, this study developed a general hierarchical classification framework for enhancing crop mapping using time-series Sentinel-1 data. Central to this method is a deep-learning-based hierarchical classification model that explores and makes use of crop hierarchical knowledge. First, preprocessed Sentinel-1 data were geometrically overlaid onto farmland parcel maps to derive parcel-scale time-series features. Second, we constructed a hierarchical crop type system for study areas based on the crop phenology of labeled crop-type samples. Third, we developed a deep-learning-based hierarchical classification model to identify crop types for each parcel, to generate final crop-type classification maps. The proposed approach was further discussed and verified through the implementation of parcel-scale time-series crop hierarchical classifications in a study area in France with farmland parcel maps and time-series Sentinel-1 data. The classification results, indicating significant improvements greater than 4.0% in overall accuracy and 5.0% in F1 score over comparative methods, demonstrated the effectiveness of the proposed method in learning multi-scale time-series features for hierarchical crop classification utilizing Sentinel-1 data sequences.

Exploiting Superpixel-Based Contextual Information on Active Learning for High Spatial Resolution Remote Sensing Image Classification

Superpixel-based classification using Active Learning (AL) has shown great potential in high spatial resolution remote sensing image classification tasks. However, in existing superpixel-based classification models using AL, the expert labeling information is only used on the selected informative superpixel while its neighboring superpixels are ignored. Actually, as most superpixels are over-segmented, a ground object always contains multiple superpixels. Thus, the center superpixel tends to have the same label as its neighboring superpixels. In this paper, to make full use of the expert labeling information, a Similar Neighboring Superpixels Search and Labeling (SNSSL) method was proposed and used in the AL process. Firstly, we identify superpixels with certain categories and uncertain superpixels by supervised learning. Secondly, we use the active learning method to process those uncertain superpixels. In each round of AL, the expert labeling information is not only used to enrich the training set but also used to label the similar neighboring superpixels. Similar neighboring superpixels are determined by computing the similarity of two superpixels according to CIELAB Dominant Colors distance, Correlation distance, Angular Second Moment distance and Contrast distance. The final classification map is composed of the supervised learning classification map and the active learning with SNSSL classification map. To demonstrate the performance of the proposed SNSSL method, the experiments were conducted on images from two benchmark high spatial resolution remote sensing datasets. The experiment shows that overall accuracy, average accuracy and kappa coefficients of the classification using the SNSSL have been improved obviously compared with the classification without the SNSSL.

Novel Adaptive Region Spectral–Spatial Features for Land Cover Classification With High Spatial Resolution Remotely Sensed Imagery

Spectral-spatial features are important for ground target identification and classification with High Spatial Resolution Remotely Sensed (HSRRS) Imagery. In this paper, two novel features, named the Gaussian-Weighting Spectral (GWS) feature and the Area Shape Index (ASI) feature, are proposed to complement the deficiency of the basic image feature for land cover classification with HSRRS imagery. The proposed GWS feature is an adaptive region-based feature that aims to improve the spectral homogeneity of a local area surrounding a pixel. Additionally, it is well known that the spectral feature is inadequate for classifying HSRRS imagery. Therefore, one spatial feature called the ASI feature is proposed here to describe the relationship between the area and shape for an adaptive region around each pixel. The proposed GWS and ASI features coupled with the basic red-green-blue feature are fed into a supervised classifier to obtain the final classification map. Experiments based on four real HSRRS images demonstrate that the proposed GWS and ASI features are capable of improving classification accuracies compared with some cognate state of the art methods. Moreover, the experiments also reveal that the proposed spectral-spatial features can complement each other for enhancing the classification performance with HSRRS images.

Two-step discriminant analysis based multi-view polarimetric SAR image classification with high confidence

Polarimetric synthetic aperture radar (PolSAR) image classification is a hot topic in remote sensing field. Although recently many deep learning methods such as convolutional based networks have provided great success in PolSAR image classification, but they need a high volume of labeled samples, which are not usually available in practice, or they cause a high computational burden for implementation. In this work, instead of spending cost for network training, the inherent nature of PolSAR image is used for generation of convolutional kernels for extraction of deep and robust features. Moreover, extraction of diverse scattering characteristics contained in the coherency matrix of PolSAR and fusion of their output classification results with a high confidence have high impact in providing a reliable classification map. The introduced method called discriminative features based high confidence classification (DFC) utilizes several approaches to deal with difficulties of PolSAR image classification. It uses a multi-view analysis to generate diverse classification maps with different information. It extracts deep polarimetric-spatial features, consistent and robust with respect to the original PolSAR data, by applying several pre-determined convolutional filters selected from the important regions of image. Convolutional kernels are fixed without requirement to be learned. The important regions are determined with selecting the key points of image. In addition, a two-step discriminant analysis method is proposed to reduce dimensionality and result in a feature space with minimum overlapping and maximum class separability. Eventually, a high confidence decision fusion is implemented to find the final classification map. Impact of multi-view analysis, selection of important regions as fixed convolutional kernels, two-step discriminant analysis and high confidence decision fusion are individually assessed on three real PolSAR images in different sizes of training sets. For example, the proposed method achieves 96.40% and 98.72% overall classification accuracy by using 10 and 100 training samples per class, respectively in L-band Flevoland image acquired by AIRSAR. Generally, the experiments show high efficiency of DFC compared to several state-of-the-art methods especially for small sample size situations.

Mapping terrestrial groundwater‐dependent ecosystems in arid Australia using Landsat‐8 time‐series data and singular value decomposition

AbstractThe spatial extent of terrestrial vegetation types reliant on groundwater in arid Australia is poorly known, largely because they are located in remote areas that are expensive to survey. In previous attempts, the use of traditional remote sensing approaches failed to discriminate vegetation using groundwater from surrounding vegetation. Difficulties in discerning vegetation groundwater use by remote sensing may be exacerbated by the unpredictable rainfall patterns and lack of annual wet and dry seasons common in arid Australia. This study presents a novel approach to mapping terrestrial groundwater‐dependent ecosystems (GDEs) by applying singular value decomposition (SVD) to time‐series of vegetation indices derived from Landsat‐8 data, to isolate the temporal and spatial sources of variation associated with groundwater use. In‐situ data from 442 sites were used to supervise and validate logistic regression models and neural networks, to determine whether sites could be correctly classified as GDEs using components obtained from the SVD. These results were used to produce a probability map of GDE occurrence across a 557 000 ha study area. Overall accuracy of the final classification map was 79%, with 72% of sites correctly identified as GDEs (true positives) and 16% incorrectly classified as GDEs (false positives). The approach is broadly applicable in arid regions globally, and is easily validated if general background knowledge of regional vegetation exists. Globally, and going forward, increased water extraction is expected to severely limit water available for GDEs. Successfully mapping GDEs in arid environments is a critical step towards their sustainable management, and the human and natural systems reliant upon them.

Nearest polarimetric and spatial neighbours for feature space projection and guidance image-based spatial filtering for PolSAR image classification

ABSTRACT A feature space projection based on the nearest polarimetric and spatial neighbours is proposed for polarimetric synthetic aperture radar (PolSAR) images. The objective function is designed such that increases the class discrimination while preserves the contextual properties. Each polarimetric or spatial neighbour is contributed proportional to with its Wishart distance to the given pixel. A guidance image, which involves all polarimetric channels, is also introduced for spatial filtering of the initial classification map, which is efficient for improving the final classification map. The proposed method shows high performance for PolSAR image classification using limited training samples.

Incremental Dictionary Learning-Driven Tensor Low-Rank and Sparse Representation for Hyperspectral Image Classification

Low-rank and sparse representation (LRSR) has gained popularity in hyperspectral image (HSI) classification. However, existing LRSR models usually treat HSI as a two-dimensional matrix, which may destroy the original 3D intrinsic structure of HSI. Moreover, the dictionary consisting of only training samples lacks completeness and may be suboptimal for representation. To overcome the above issues, we propose an incremental dictionary learning-driven tensor low-rank and sparse representation (TLRSR-IDL) model for HSI classification. First, we represent HSI as a third-order tensor to retain its original 3D intrinsic structure by using the TLRSR model, which also combines both sparsity and low rankness to maintain global and local data structures. Second, we design an optimal reconstruction within regularized neighborhood (ORRN) method to exploit spectral-spatial information by avoiding the interference of heterogeneous samples in the neighborhood. Finally, an incremental dictionary learning (IDL) scheme is designed to iteratively introduce augmented samples into the dictionary, and the final classification map is produced by feeding back the last round of the incremental dictionary into the TLRSR-IDL model. The main innovative contribution lies in that the proposed IDL scheme can leverage supervised and unsupervised information, which greatly enhances traditional LRSR and TLRSR models. Experimental results based on three popular hyperspectral datasets demonstrate that the proposed method outperforms other related counterparts in terms of classification accuracy and generalization performance, with OA improvements of 0.97%-16.83%, 1.25%-6.89%, and 0.85%-6.67% for Indian Pines, Pavia University, and Salinas, respectively.

Characterizing the Up-To-Date Land-Use and Land-Cover Change in Xiong’an New Area from 2017 to 2020 Using the Multi-Temporal Sentinel-2 Images on Google Earth Engine

Land use and land cover (LULC) are fundamental units of human activities. Therefore, it is of significance to accurately and in a timely manner obtain the LULC maps where dramatic LULC changes are undergoing. Since 2017 April, a new state-level area, Xiong’an New Area, was established in China. In order to better characterize the LULC changes in Xiong’an New Area, this study makes full use of the multi-temporal 10-m Sentinel-2 images, the cloud-computing Google Earth Engine (GEE) platform, and the powerful classification capability of random forest (RF) models to generate the continuous LULC maps from 2017 to 2020. To do so, a novel multiple RF-based classification framework is adopted by outputting the classification probability based on each monthly composite and aggregating the multiple probability maps to generate the final classification map. Based on the obtained LULC maps, this study analyzes the spatio-temporal changes of LULC types in the last four years and the different change patterns in three counties. Experimental results indicate that the derived LULC maps achieve high accuracy for each year, with the overall accuracy and Kappa values no less than 0.95. It is also found that the changed areas account for nearly 36%, and the dry farmland, impervious surface, and other land-cover types have changed dramatically and present varying change patterns in three counties, which might be caused by the latest planning of Xiong’an New Area. The obtained 10-m four-year LULC maps in this study are supposed to provide some valuable information on the monitoring and understanding of what kinds of LULC changes have taken place in Xiong’an New Area.

A random patches based edge preserving network for land cover classification using Polarimetric Synthetic Aperture Radar images

ABSTRACT A random patches-based edge-preserving network (RPEP) is proposed for polarimetric synthetic aperture radar (PolSAR) image classification in this paper. An initial spatial feature extraction is firstly done using the transform domain recursive filtering. From the filtered images, several random patches are chosen and used as convolutional kernels. The designed random patches-based network uses these fix kernels without doing any training process. The multi-scale features extracted by both shallow and deep layers are given to a support vector machine to get an initial classification map. The binary probability maps obtained from the initial classification map are then smoothed using the guided filter as an edge preserving filter. The final classification map is achieved by applying the maximum decision rule. The proposed RPEP method with extraction of robust, consistent, invariant and multi-scale polarimetric-spatial features and also by doing noise reduction with edge preserving filters provides superior classification results compared to several state-of-the-art methods especially in small sample size situations. In addition, RPEP has a simple and fast implementation that makes it a powerful classifier for real applications.

Adaptive region-based post-classification framework for land-cover mapping improvement using very high spatial resolution optical imagery

Land cover mapping using remote sensing optical images with a very high spatial resolution (VHSR) plays an important role in observing the Earth’s surface. However, classification maps are usually affected by salt-and-pepper noise because VHSR optical images usually have a low resolution of ground targets in terms of spectral reflectance. An adaptive region-based post-classification framework (ARPF) is proposed for improving the initial classification map while using a VHSR optical image to further smooth the noise of initial classified maps. First, different from several traditional methods using a single classifier, our proposed ARPF needs more than four different initial classification maps acquired from different classifiers or image features. Second, an adaptive region around each pixel of the gray image is generated with two predefined parameters, and each adaptive region is applied to refine the corresponding pixel of each initial classified map. Finally, all the refined classified maps are merged to obtain the final classification map by coupling adaptive region and majority voting rules. In our experiments, three optical images with VHSR are used to evaluate the proposed ARPF. Compared with three typical relevant post-classification methods, the proposed ARPF can provide a classification map with less noise in visual performance and achieve higher quantitative accuracy while having an advantage in the constant detail of ground targets.

Multiscale filter-based hyperspectral image classification with PCA and SVM

Abstract Hyperspectral imagery can offer images with high spectral resolution and provide a unique ability to distinguish the subtle spectral signatures of different land covers. In this paper, we develop a new algorithm for hyperspectral image classification by using principal component analysis (PCA) and support vector machines (SVM). We use PCA to reduce the dimensionality of an HSI data cube, and then perform spatial convolution with three different filters on the PCA output cube. We feed all three convolved output cubes to SVM to classify every pixel. Finally, we perform fusion on the three output maps to determine the final classification map. We conduct experiments on three widely used hyperspectral image data cubes (ie indian pines, pavia university, and salinas). Our method can improve the classification accuracy significantly when compared to several existing methods. Our novel method is relatively fast in term of CPU computational time as well.

Remote sensing image classification using subspace sensor fusion

The amount of remote sensing and ancillary datasets captured by diverse airborne and spaceborne sensors has been tremendously increased, which opens up the possibility of utilizing multimodal datasets to improve the performance of processing approaches with respect to the application at hand. However, developing a generic framework with high generalization capability that can effectively fuse diverse datasets is a challenging task since the current approaches are usually only applicable to two specific sensors for data fusion. In this paper, we propose an accurate fusion-based technique called SubFus with capability to integrate diverse remote sensing data for land cover classification. Here, we assume that a high dimensional multisensor dataset can be represented fused features that live in a lower-dimensional space. The proposed classification methodology includes three main stages. First, spatial information is extracted by using spatial filters (i.e., morphology filters). Then, a novel low-rank minimization problem is proposed to represent the multisensor datasets in subspaces using fused features. The fused features in the lower-dimensional subspace are estimated using a novel iterative algorithm based on the alternative direction method of multipliers. Third, the final classification map is produced by applying a supervised spectral classifier (i.e., random forest) on the fused features. In the experiments, the proposed method is applied to a three-sensor (RGB, multispectral LiDAR, and hyperspectral images) dataset captured over the area of the University of Houston, the USA, and a two-sensor (hyperspectral and LiDAR) dataset captured over the city of Trento, Italy. The land-cover maps generated using SubFus are evaluated based on classification accuracies. Experimental results obtained by SubFus confirm considerable improvements in terms of classification accuracies compared with the other methods used in the experiments. The proposed fusion approach obtains 85.32% and 99.25% in terms of overall classification accuracy on the Houston (the training portion of the dataset distributed for the data fusion contest of 2018) and trento datasets, respectively.

Local Binary Ensemble based Self-training for Semi-supervised Classification of Hyperspectral Remote Sensing Images

Supervised classification of hyperspectral remote sensing images is still challenging due to the scarcity of enough labelled samples. Semi-supervised methods have been adopted to handle this issue. Self-training is a popular semi-supervised technique which is widely used for training a classifier with limited labelled data and a large quantity of unlabeled data. However, traditional self-training approaches often give poor classification results in high dimensional data. In the current work, a novel efficient self-training approach for handling the deficiency of labelled samples for semi-supervised classification of hyperspectral remote sensing images is proposed. The proposed method first trains an ensemble of locally specialized supervised binary classifiers independently by using the dimensionally reduced spectral feature vectors of few available labelled samples. The trained local binary classifiers are then used to extend the labelled set by iterative addition of highly informative unlabeled samples to it by exploiting both the spectral and spatial information of the hyperspectral image. The classifiers are then retrained with the extended dataset in a batchwise manner and the procedure is repeated until adequate quantity of labelled samples are generated. Finally, a supervised multiclass classifier is trained on the extended dataset to produce the final classification map. Experimental results on two benchmark hyperspectral image datasets prove the effectiveness of the proposed method over supervised and traditional self-training based semi-supervised pixelwise classification approach in terms of different classification measures.

A novel spectral–spatial based adaptive minimum spanning forest for hyperspectral image classification

The classification methods based on minimum spanning forest (MSF) have yielded impressive results for hyperspectral image. However, previous methods exist several drawbacks, i.e., marker selection methods are easily affected by boundary noise pixels, dissimilarity measure methods between pixels are inaccurate, and also image segmentation process is not robust, since they have not effectively utilized spatial information. To this end, in this paper, novel gradient-based marker selection technique, dissimilarity measures, and adaptive connection weighting method are proposed by making full use of spatial information in hyperspectral image. Concretely, for a given hyperspectral image, a pixel-wise classification is firstly performed, and meanwhile the gradient map is generated by a morphology-based algorithm. Secondly, the most reliable pixels are selected as the markers from the classification map, and then the boundary noise pixels are excluded from the marker map by using the gradient map. Thirdly, several new dissimilarity measures are proposed by incorporating gradient information or probability information of pixels. Furthermore, in the growth procedure of MSF, the connection weighting between pixels is adjusted adaptively to improve the robustness of the MSF algorithm. Finally, when building the final classification map by using the majority voting rule, the labels of the training samples are used to dominate the label prediction. Experimental results are performed on two hyperspectral image sets Indian Pines and University of Pavia with different resolutions and contexts. The proposed approach yields higher classification accuracies compared to previously proposed classification methods, and provides accurate segmentation maps.

An overview on spectral and spatial information fusion for hyperspectral image classification: Current trends and challenges

Hyperspectral images (HSIs) have a cube form containing spatial information in two dimensions and rich spectral information in the third one. The high volume of spectral bands allows discrimination between various materials with high details. Moreover, by utilizing the spatial features of image such as shape, texture and geometrical structures, the land cover discrimination will be improved. So, fusion of spectral and spatial information can significantly improve the HSI classification. In this work, the spectral-spatial information fusion methods are categorized into three main groups. The first group contains segmentation based methods where objects or super-pixels are used instead of pixels for classification or the obtained segmentation map is used for relaxation of the pixel-wise classification map. The second group consists of feature fusion methods which are divided into six sub-groups: features stacking, joint spectral-spatial feature extraction, kernel based classifiers, representation based classifiers, 3D spectral-spatial feature extraction and deep learning based classifiers. The third fusion methods are decision fusion based approaches where complementary information of several classifiers are contributed for achieving the final classification map. A review of different methods in each category, is presented. Moreover, the advantages and difficulties/disadvantages of each group are discussed. The performance of various fusion methods are assessed in terms of classification accuracy and running time using experiments on three popular hyperspectral images. The results show that the feature fusion methods although are time consuming but can provide superior classification accuracy compared to other methods. Study of this work can be very useful for all researchers interested in HSI feature extraction, fusion and classification.

Multiple Kernel-Based SVM Classification of Hyperspectral Images by Combining Spectral, Spatial, and Semantic Information

In this study, we present a hyperspectral image classification method by combining spectral, spatial, and semantic information. The main steps of the proposed method are summarized as follows: First, principal component analysis transform is conducted on an original image to produce its extended morphological profile, Gabor features, and superpixel-based segmentation map. To model spatial information, the extended morphological profile and Gabor features are used to represent structure and texture features, respectively. Moreover, the mean filtering is performed within each superpixel to maintain the homogeneity of the spatial features. Then, the k-means clustering and the entropy rate superpixel segmentation are combined to produce semantic feature vectors by using a bag of visual-words model for each superpixel. Next, three kernel functions are constructed to describe the spectral, spatial, and semantic information, respectively. Finally, the composite kernel technique is used to fuse all the features into a multiple kernel function that is fed into a support vector machine classifier to produce a final classification map. Experiments demonstrate that the proposed method is superior to the most popular kernel-based classification methods in terms of both visual inspection and quantitative analysis, even if only very limited training samples are available.

A Bilevel Contextual MRF Model for Supervised Classification of High Spatial Resolution Remote Sensing Images

Markov random field (MRF) based methods have been widely used in high spatial resolution (HSR) image classification. However, many existing MRF-based methods put more emphasis on pixel level contexts while less on superpixel level contextual information. To cope with this issue, this article presents a novel bilevel contextual MRF framework, named BLC-MRF, for HSR imagery classification. Specifically, pixel and superpixel level dependence are incorporated into the proposed MRF model to fully exploit spectral–spatial contextual information and preserve object boundaries in HSR images. In BLC-MRF, a pixel level MRF model is first performed and then cascaded as an input of a superpixel level MRF. In superpixel level, unary and pairwise potential terms are constructed by using the superpixel probability estimation method and spectral histogram distance, respectively. At last, a contextual MRF model is conducted and the final classification map can be computed by using $\alpha$ -expansion algorithm. The benefits of BLC-MRF are twofold: first, the pixel and superpixel level contextual information can be exploited under MRF framework to preserve object boundaries for improving the classification performance, and, second, the algorithm can provide promising results with a small number of training samples. Experimental results on three HSR datasets demonstrate that the proposed approach outperforms several state-of-the-art methods in terms of the classification performance.

Semisupervised Classification Based on SLIC Segmentation for Hyperspectral Image

With the high spectral resolution, hyperspectral image (HSI) can provide a wealth of information for image classification. Many classification methods utilize the training samples to classify the ground materials. However, the small sample problem is still urgent to be solved when considering the cost of labeling training samples. In order to solve this problem, this letter proposes a semisupervised classification method based on the simple linear iterative cluster (SLIC) segmentation for HSI. This method improves the SLIC method to better explore the spectral characteristic of HSI. It explores the learned superpixel map and initial classification map to select the pseudo-labeled samples (PLSs), which is expected to increase the effectiveness of PLSs. The final classification map can be obtained with the integrated labeled training samples and PLSs. Experiments were carried out on three HSIs, and it was founded that the proposed method generally shows a better classification performance than the other methods.

Novel Multi-Scale Filter Profile-Based Framework for VHR Remote Sensing Image Classification

Filter is a well-known tool for noise reduction of very high spatial resolution (VHR) remote sensing images. However, a single-scale filter usually demonstrates limitations in covering various targets with different sizes and shapes in a given image scene. A novel method called multi-scale filter profile (MFP)-based framework (MFPF) is introduced in this study to improve the classification performance of a remote sensing image of VHR and address the aforementioned problem. First, an adaptive filter is extended with a series of parameters for MFP construction. Then, a layer-stacking technique is used to concatenate the MPFs and all the features into a stacked vector. Afterward, principal component analysis, a classical descending dimension algorithm, is performed on the fused profiles to reduce the redundancy of the stacked vector. Finally, the spatial adaptive region of each filter in the MFPs is used for post-processing of the obtained initial classification map through a supervised classifier. This process aims to revise the initial classification map and generate a final classification map. Experimental results performed on the three real VHR remote sensing images demonstrate the effectiveness of the proposed MFPF in comparison with the state-of-the-art methods. Hard-tuning parameters are unnecessary in the application of the proposed approach. Thus, such a method can be conveniently applied in real applications.

DCN-Based Spatial Features for Improving Parcel-Based Crop Classification Using High-Resolution Optical Images and Multi-Temporal SAR Data

Spatial features retrieved from satellite data play an important role for improving crop classification. In this study, we proposed a deep-learning-based time-series analysis method to extract and organize spatial features to improve parcel-based crop classification using high-resolution optical images and multi-temporal synthetic aperture radar (SAR) data. Central to this method is the use of multiple deep convolutional networks (DCNs) to extract spatial features and to use the long short-term memory (LSTM) network to organize spatial features. First, a precise farmland parcel map was delineated from optical images. Second, hundreds of spatial features were retrieved using multiple DCNs from preprocessed SAR images and overlaid onto the parcel map to construct multivariate time-series of crop growth for parcels. Third, LSTM-based network structures for organizing these time-series features were constructed to produce a final parcel-based classification map. The method was applied to a dataset of high-resolution ZY-3 optical images and multi-temporal Sentinel-1A SAR data to classify crop types in the Hunan Province of China. The classification results, showing an improvement of greater than 5.0% in overall accuracy relative to methods without spatial features, demonstrated the effectiveness of the proposed method in extracting and organizing spatial features for improving parcel-based crop classification.