AVHRR channel selection for land cover classification

Mapping land cover of large regions often requires processing of satellite images collected from several time periods at many spectral wavelength channels. However, manipulating and processing large amounts of image data increases the complexity and time, and hence the cost, that it takes to produce a land cover map. Very few studies have evaluated the importance of individual Advanced Very High Resolution Radiometer (AVHRR) channels for discriminating cover types, especially the thermal channels (channels 3, 4 and 5). Studies rarely perform a multi-year analysis to determine the impact of inter-annual variability on the classification results. We evaluated 5 years of AVHRR data using combinations of the original AVHRR spectral channels (1-5) to determine which channels are most important for cover type discrimination, yet stabilize inter-annual variability. Particular attention was placed on the channels in the thermal portion of the spectrum. Fourteen cover types over the entire state of Colorado were evaluated using a supervised classification approach on all two-, three-, four- and five-channel combinations for seven AVHRR biweekly composite datasets covering the entire growing season for each of 5 years. Results show that all three of the major portions of the electromagnetic spectrum represented by the AVHRR sensor are required to discriminate cover types effectively and stabilize inter-annual variability. Of the two-channel combinations, channels 1 (red visible) and 2 (near-infrared) had, by far, the highest average overall accuracy (72.2%), yet the inter-annual classification accuracies were highly variable. Including a thermal channel (channel 4) significantly increased the average overall classification accuracy by 5.5% and stabilized interannual variability. Each of the thermal channels gave similar classification accuracies; however, because of the problems in consistently interpreting channel 3 data, either channel 4 or 5 was found to be a more appropriate choice. Substituting the thermal channel with a single elevation layer resulted in equivalent classification accuracies and inter-annual variability.     

Object-based land cover classification using airborne LiDAR

Light Detection and Ranging (LiDAR) provides high resolution horizontal and vertical spatial point cloud data, and is increasingly being used in a number of applications and disciplines, which have concentrated on the exploit and manipulation of the data using mainly its three dimensional nature. LiDAR information potential is made even greater though, with its consideration of intensity.Elevation and intensity airborne LiDAR data are used in this study in order to classify forest and ground types quickly and efficiently without the need for manipulating multispectral image files, using a supervised object-orientated approach. LiDAR has the advantage of being able to create elevation surfaces that are in 3D, while also having information on LiDAR intensity values, thus it is a spatial and spectral segmentation tool. This classification method also uses point distribution frequency criteria to differentiate between land cover types. Classifications were performed using two methods, one that included the influence of the ground in heavily vegetated areas, and the other which eliminated the ground points before classification. The classification of three meanders of the Garonne and Allier rivers in France has demonstrated overall classification accuracies of 95% and 94% for the methods including and excluding the ground influence respectively. Five types of riparian forest were classified with accuracies between 66 and 98%. These forest types included planted and natural forest stands of different ages. Classifications of short vegetation and bare earth also produced high accuracies averaging above 90%.     

Sub-pixel land cover mapping for per-field classification

A method was developed to transform a soft land cover classification into hard land cover classes at the sub-pixel scale for subsequent per-field classification. First, image pixels were segmented using vector boundaries. Second, the pixel segments (ranked by area) were labelled with a land cover class (ranked by class typicality). Third, a hard per-field classification was generated by examining each polygon (representing a land cover parcel, or field) in its entirety (by grouping the fragments of the polygon contained within different image pixels) and assigning to it the modal land cover class. The accuracy of this technique was considerably higher than that of both a corresponding hard per-pixel classification and a perfield classification based on hard per-pixel classified imagery.     

AVHRR composite period selection for land cover classification

Multitemporal satellite image datasets provide valuable information on the phenological characteristics of vegetation, thereby significantly increasing the accuracy of cover type classifications compared to single date classifications. However, the processing of these datasets can become very complex when dealing with multitemporal data combined with multispectral data. Advanced Very High Resolution Radiometer (AVHRR) biweekly composite data are commonly used to classify land cover over large regions. Selecting a subset of these biweekly composite periods may be required to reduce the complexity and cost of land cover mapping. The objective of our research was to evaluate the effect of reducing the number of composite periods and altering the spacing of those composite periods on classification accuracy. Because inter-annual variability can have a major impact on classification results, 5 years of AVHRR data were evaluated. AVHRR biweekly composite images for spectral channels 1-4 (visible, nearinfrared and two thermal bands) covering the entire growing season were used to classify 14 cover types over the entire state of Colorado for each of five different years. A supervised classification method was applied to maintain consistent procedures for each case tested. Results indicate that the number of composite periods can be halved-reduced from 14 composite dates to seven composite dates-without significantly reducing overall classification accuracy (80.4% Kappa accuracy for the 14-composite dataset as compared to 80.0% for a seven-composite dataset). At least seven composite periods were required to ensure the classification accuracy was not affected by inter-annual variability due to climate fluctuations. Concentrating more composites near the beginning and end of the growing season, as compared to using evenly spaced time periods, consistently produced slightly higher classification values over the 5 years tested (average Kappa of 80.3% for the heavy early/late case as compared to 79.0% for the alternate dataset case).     

Improving Thematic Mapper land cover classification using filtered data

In an attempt to alleviate the classification problems introduced by the higher spatial resolution of the Thematic Mapper in comparison to the Muitispectral Scanner, classifications were performed on two to six band combinations, first using Thematic Mapper bands only, and subsequently replacing band 5 by its mean-filtered and median-filtered counterpart. The combination of filtered data with non-filtered data smooths out scene noise while retaining some of the boundary detail.     

An artificial immune network approach to multi-sensor land use/land cover classification

An optimized artificial immune network-based classification model, namely OPTINC, was developed for remote sensing-based land use/land cover (LULC) classification. Major improvements of OPTINC compared to a typical immune network-based classification model (aiNet) include (1) preservation of the best antibodies of each land cover class from the antibody population suppression, which ensures that each land cover class is represented by at least one antibody; (2) mutation rates being self-adaptive according to the model performance between training generations, which improves the model convergence; and (3) incorporation of both Euclidean distance and spectral angle mapping distance to measure affinity between two feature vectors using a genetic algorithm-based optimization, which helps the model to better discriminate LULC classes with similar characteristics. OPTINC was evaluated using two sites with different remote sensing data: a residential area in Denver, CO with high-spatial resolution QuickBird image and LiDAR data, and a suburban area in Monticello, UT with HyMap hyperspectral imagery. A decision tree, a multilayer feed-forward back-propagation neural network, and aiNet were also tested for comparison. Classification accuracy, local homogeneity of classified images, and model sensitivity to training sample size were examined. OPTINC outperformed the other models with higher accuracy and more spatially cohesive land cover classes with limited salt-and-pepper noise. OPTINC was relatively less sensitive to training sample size than the neural network, followed by the decision tree.     

Boosting a fast neural network for supervised land cover classification

It is demonstrated that the use of an ensemble of neural networks for routine land cover classification of multispectral satellite data can lead to a significant improvement in classification accuracy. Specifically, the AdaBoost.M1 algorithm is applied to a sequence of three-layer, feed-forward neural networks. In order to overcome the drawback of long training time for each network in the ensemble, the networks are trained with an efficient Kalman filter algorithm. On the basis of statistical hypothesis tests, classification performance on multispectral imagery is compared with that of maximum likelihood and support vector machine classifiers. Good generalization accuracies are obtained with computation times of the order of 1 h or less. The algorithms involved are described in detail and a software implementation in the ENVI/IDL image analysis environment is provided.     

Effect of Landsat thematic mapper sensor parameters on land cover classification

Selected sensor parameter differences between TM and MSS were assessed through classification performance of a suburban/regional test site. Overall classification accuracy of a seven-band Landsat TM scene in comparison to MSS yielded an improvement in accuracy from 74.8% to 83.2%. To study the possible causes for the difference in classification performance, key sensor parameter differences between MSS and TM, including 1) spatial resolution (30 m for TM versus 80 m for MSS), 2) quantization level (256 levels for TM versus 64 for MSS), and 3) spectral regions (seven bands in four major spectral regions for TM versus four bands in two regions for MSS), were evaluated. Landsat TM data were processed to simulate all possible combinations of these MSS and TM parameters, yielding a three-factor design with two levels per factor. The results indicated that the added spectral regions (TM 1, TM 5, and TM 7) and to a lesser degree the increase in quantization level to eight bits produced the improved TM classification accuracy. However, in this study, the higher 30 m spatial resolution of TM contributed to a reduced classification accuracy from increased within-field variability or class heterogeneity.     

Scanning,compression and land cover classification of astronaut-acquired orbital photographs

Studies that utilize astronaut-acquired orbital photographs for visual or digital classification require high-quality data to ensure accuracy. The majority of images available must be scanned from film and electronically transferred to scientific users. This study examined the effect of scanning spatial resolution (1200, 2400?pixels per inch (21.2 and 10.6?µm/pixel)), scanning density range option (Auto, Full) and compression ratio (non-lossy: Tagged-Image File Format (TIFF); and lossy: Joint Photographic Experts Group (JPEG) 10:1, 46:1, 83:1) on digital classification results of an orbital photograph from the National Aeronautics and Space Administration (NASA)–Johnson Space Center archive. Qualitative results suggested that 1200?ppi was acceptable for visual interpretative uses for major land cover types. Moreover, Auto scanning density range was superior to Full density range. Quantitative assessment of the processing steps indicated that, while 2400?ppi scanning spatial resolution resulted in more classified polygons as well as a substantially greater proportion of polygons ≤0.2?ha, overall agreement between 1200?ppi and 2400?ppi was quite high. JPEG compression up to approximately 46:1 also did not appear to have a major impact on quantitative classification characteristics. We conclude that both 1200 and 2400?ppi scanning resolutions are acceptable options for this level of land cover classification, as well as a compression ratio at or below approximately 46:1. Auto range density should always be used during scanning because it acquires more of the information from the film. The particular combination of scanning spatial resolution and compression level will require a case-by-case decision and will depend upon memory capabilities, analytical objectives and the spatial properties of the objects in the image.     

Global land cover classification at 1 km spatial resolution using a classification tree approach

This paper on reports the production of a 1 km spatial resolution land cover classification using data for 1992-1993 from the Advanced Very High Resolution Radiometer (AVHRR). This map will be included as an at-launch product of the Moderate Resolution Imaging Spectroradiometer (MODIS) to serve as an input for several algorithms requiring knowledge of land cover type. The methodology was derived from a similar effort to create a product at 8 km spatial resolution, where high resolution data sets were interpreted in order to derive a coarse-resolution training data set. A set of 37 294 x 1 km pixels was used within a hierarchical tree structure to classify the AVHRR data into 12 classes. The approach taken involved a hierarchy of pair-wise class trees where a logic based on vegetation form was applied until all classes were depicted. Multitemporal AVHRR metrics were used to predict class memberships. Minimum annual red reflectance, peak annual Normalized Difference Vegetation Index (NDVI), and minimum channel three brightness temperature were among the most used metrics. Depictions of forests and woodlands, and areas of mechanized agriculture are in general agreement with other sources of information, while classes such as low biomass agriculture and high-latitude broadleaf forest are not. Comparisons of the final product with regional digital land cover maps derived from high-resolution remotely sensed data reveal general agreement, except for apparently poor depictions of temperate pastures within areas of agriculture. Distinguishing between forest and non-forest was achieved with agreements ranging from 81 to 92% for these regional subsets. The agreements for all classes varied from an average of 65% when viewing all pixels to an average of 82% when viewing only those 1 km pixels consisting of greater than 90% one class within the high-resolution data sets.     

Advances in classification for land cover mapping using SPOT HRV imagery

Abstract

High-resolution data from the HRV (High Resolution Visible) sensors onboard the SPOT-1 satellite have been utilized for mapping semi-natural and agricultural land cover using automated digital image classification algorithms. Two methods for improving classification performance are discussed. The first technique involves the use of digital terrain information to reduce the effects of topography on spectral information while the second technique involves the classification of land-cover types using training data derived from spectral feature space. Test areas in Snowdonia and the Somerset Levels were used to evaluate the methodology and promising results were achieved. However, the low classification accuracies obtained suggest that spectral classification alone is not a suitable tool to use in the mapping of semi-natural cover types.     

Performance analysis of IRS-bands for land use/land cover classification system using Maximum Likelihood Classifier

The allocation of confidence level to a classification product is considered to be essential. The acquisition of site-specific data to check the classification was discussed. A statistical approach to the determination of an appropriate confidence level from the check data was presented. The test scene was classified using Maximum likelihood classification with various band combinations. These combinations were used as statistical variables. The results are analysed and the best possible combinations selected for accurate classification system. Keeping the percentage overall accuracy in view, these tests demonstrated the complexity of relations between the land cover classes and the data processing variables. Further, they indicated the variables best suited to the classification of certain classes, based on the performance of variable by class. Here, Hyderabad City is taken as a test site.     

Mapping urban land cover based on spatial-spectral classification of hyperspectral remote-sensing data

In this article, an innovative classification framework for hyperspectral image data, based on both spectral and spatial information, is proposed. The main objective of this method is to improve the accuracy and efficiency of high-resolution land-cover mapping in urban areas. The spatial information is obtained by an enhanced marker-based minimum spanning forest (MMSF) algorithm. A pixel-based support vector machine (SVM) algorithm is first used to classify the hyperspectral image data, then the enhanced MMSF algorithm is applied in order to increase the accuracy of less accurately classified land-cover types. The enhanced MMSF algorithm is used as a binary classifier. These two classes are the low-accuracy class and remaining classes. Finally, the SVM algorithm is trained for classes with acceptable accuracy. In the proposed approach, namely MSF-SVM, the markers are extracted from the classification maps obtained by both SVM and watershed segmentation algorithms, and are then used to build the MSF. Three benchmark hyperspectral data sets are used for the assessment: Berlin, Washington DC Mall, and Quebec City. Experimental results demonstrate the superiority of the proposed approach compared with SVM and the original MMSF algorithms. It achieves approximately 5, 6, and 7% higher rates in kappa coefficients of agreement in comparison with the original MMSF algorithm for the Berlin, Washington DC Mall, and Quebec City data sets, respectively.     

Fourier analysis of multi-temporal AVHRR data applied to a land cover classification

A signal processing technique is presented and applied to annual patterns of the Global Vegetation Index (GVI) derived from the Advanced Very High Resolution Radiometer (AVHRR) to examine the frequency distribution of the multi-temporal signal. It is shown that frequencies of the signal are linked to integrated GVI, seasonal variability and subseasonal variability of the land cover type. These characteristics are used to derive a land cover classification.     

Characterization and classification of South American land cover types using satellite data

Abstract

Various methods are compared for carrying out land cover classifications of South America using multitemporal Advanced Very High Resolution Radiometer data. Fifty-two images of the normalized difference vegetation index (NDVI) from a 1-year period are used to generate multitemporal data sets. Three main approaches to land cover classification are considered, namely the use of the principal components transformed images, the use of a characteristic curves procedure based on N DVI values plotted against time, and finally application of the maximum likelihood rule to multitemporal data sets. Comparison of results from training sites indicates that the last approach yields the most accurate results. Despite the reliance on training site figures for performance assessment, the results are nevertheless extremely encouraging, with accuracies for several cover types exceeding 90 per cent.     

A novel algorithm for land use and land cover classification using RADARSAT-2 polarimetric SAR data

This study proposes a new four-component algorithm for land use and land cover (LULC) classification using RADARSAT-2 polarimetric SAR (PolSAR) data. These four components are polarimetric decomposition, PolSAR interferometry, object-oriented image analysis, and decision tree algorithms. First, polarimetric decomposition can be used to support the classification of PolSAR data. It is aimed at extracting polarimetric parameters related to the physical scattering mechanisms of the observed objects. Second, PolSAR interferometry is used to extract polarimetric interferometric information to support LULC classification. Third, the main purposes of object-oriented image analysis are delineating image objects, as well as extracting various textural and spatial features from image objects to improve classification accuracy. Finally, a decision tree algorithm provides an efficient way to select features and implement classification. A comparison between the proposed method and the Wishart supervised classification which is based on the coherency matrix was made to test the performance of the proposed method. The overall accuracy of the proposed method was 86.64%, whereas that of the Wishart supervised classification was 69.66%. The kappa value of the proposed method was 0.84, much higher than that of the Wishart supervised classification, which exhibited a kappa value of 0.65. The results indicate that the proposed method exhibits much better performance than the Wishart supervised classification for LULC classification. Further investigation was carried out on the respective contribution of the four components to LULC classification using RADARSAT-2 PolSAR data, and it indicates that all the four components have important contribution to the classification. Polarimetric information has significant implications for identifying different vegetation types and distinguishing between vegetation and urban/built-up. The polarimetric interferometric information extracted from repeat-pass RADARSAT-2 images is important in reducing the confusion between urban/built-up and vegetation and that between barren/sparsely vegetated land and vegetation. Object-oriented image analysis is very helpful in reducing the effect of speckle in PolSAR images by implementing classification based on image objects, and the textural information extracted from image objects is helpful in distinguishing between water and lawn. The decision tree algorithm can achieve higher classification accuracy than the nearest neighbor classification implemented using Definiens Developer 7.0, and the accuracy of the decision tree algorithm is similar with that of the support vector classification which is implemented based on the features selected using genetic algorithms. Compared with the nearest neighbor and support vector classification, the decision tree algorithm is more efficient to select features and implement classification. Furthermore, the decision tree algorithm can provide clear classification rules that can be easily interpreted based on the physical meaning of the features used in the classification. This can provide physical insight for LULC classification using PolSAR data.     

Topographic constrained land cover classification in mountain areas using fully convolutional network

Mountains are an important kind of landform on the earth’s surface. Due to harsh mountainous environment, such as steep slopes and cliffs, remote sensing has become an indispensable tool for surveying mountain areas instead of traditional ground surveys. However, the accuracy of current land cover products derived from remote sensing in mountain areas is still low. In this paper, we propose a three-level architecture for land cover classification in mountain areas. Topographic partitioning is first performed in order to partition a large area into several smaller zones, and then, multiresolution segmentation is implemented in each individual zone. Thus, we can obtain initial geo-semantic objects with terrain, spectrum and texture homogeneities. A fully convolutional network (FCN)-based classifier (U-Net) is further introduced for supervised classification of land cover. From the perspectives of both visual interpretation and quantitative evaluation, the proposed method achieved robust and high-precision results for all land cover types. We also investigate the contributions of multimodal features for classification accuracy improvement. First, the results showed that additional features resulted in higher classification accuracies than 3-features only; 6-features achieved the best performance on farmland, impervious surfaces and coniferous forests, while 5-features performed well on water and broad-leaved forests. The elevation feature did not have a positive effect on water and broad-leaved forests, which can be explained by their physical distribution in the landscape. Second, the most significant improvement was achieved on water (Kappa coefficient increased from 0.741 to 0.924), followed by coniferous forests (Kappa coefficient increased from 0.629 to 0.805), whereas only a minor improvement was observed for the other three types. Furthermore, the accuracies of farmland and impervious surfaces remained relatively high even without the assistance of additional features, and texture feature plays a key role. The final land cover map was generated by combining the optimal results of each type via a hierarchical integrating strategy. The overall accuracy of classification achieved 90.6%.     

Supervised land cover classification based on the locally reduced convex hull approach

A novel supervised learning approach, called the locally reduced convex hull (LRCH), is proposed for land cover classification. The method described is capable of increasing the class separability and the representational capacity of the training set, which leads to its high generalization ability in applications. The effectiveness of the LRCH is demonstrated on the classification problem of a multi-spectral data set. In experiments, the LRCH was compared with six common classifiers. Statistical results in terms of the overall accuracy, the Kappa coefficient and McNemar's test show that LRCH outperforms most of the other approaches, with a speed that is comparable to all of them.     

Reusing back-propagation artificial neural networks for land cover classification in tropical savannahs

This paper focuses on a method to overcome some of the disadvantages that are related with the use of artificial neural networks (ANNs) as supervised classifiers. The proposed method aims at speeding up network learning, improving classification accuracies and reducing variability on classification performance due to random weight initialization. This can be realized by transferring implicit knowledge from a previously learned source task to a new target task using the proposed algorithm, Discriminality Based Transfer (DBT). The presented approach is compared with conventional network training and a literal transfer method in a 13-class tropical savannah classification experiment using Landsat Thematic Mapper (TM) data. Knowledge was extracted from a network trained on the Kara experimental site in Togo. This information was used to classify the Savanes-L'Oti area which differs in terms of geographical position, image acquisition date, climatological condition and land cover. It was possible to speed up network learning 5.2, 4.3 and 1.8 times using, respectively, 5-, 10- and 20-pixels-per-class training sets. Larger training sets showed less speed improvement. After applying DBT, average classification accuracies were not significantly different from accuracies obtained after training random initialized networks, although DBT tended to show better performance on smaller training sets. It was possible to explain differences in individual class accuracies by analysing Bhattacharyya (BH) distances calculated between all Kara and Savanes-L'Oti classes. Finally, variability on classification performance decreased significantly when training with 5-, 10- and 20-pixels-per-class training sets after DBT application.     

A multistage genetic fuzzy classifier for land cover classification from satellite imagery

A linguistic boosted genetic fuzzy classifier (LiBGFC) is proposed in this paper for land cover classification from multispectral images. The LiBGFC is a three-stage process, aiming at effectively tackling the interpretability versus accuracy tradeoff problem. The first stage iteratively generates fuzzy rules, as directed by a boosting algorithm that localizes new rules in uncovered subspaces of the feature space, implicitly preserving the cooperation with previously derived ones. Each rule is able to select the required features, further improving the interpretability of the obtained model. Special provision is taken in the formulation of the fitness function to avoid the creation of redundant rules. A simplification stage follows the first one aiming at further improving the interpretability of the initial rule base, providing a more compact and interpretable solution. Finally, a genetic tuning stage fine tunes the fuzzy sets database improving the classification performance of the obtained model. The LiBGFC is tested using an IKONOS multispectral VHR image, in a lake-wetland ecosystem of international importance. The results indicate the effectiveness of the proposed system in handling multidimensional feature spaces, producing easily understandable fuzzy models.