Pixel-based Classification Results Research Articles

Fine-grained wetland classification for national wetland reserves using multi-source remote sensing data and Pixel Information Expert Engine (PIE-Engine)

ABSTRACT Timely and accurate wetland information is necessary for wetland resource management. Recent advances in machine learning and remote sensing have facilitated cost-effective monitoring of wetlands. However, reliable methods for fine-grained and rapid wetland mapping are still lacking. To address the issue, a wetland sample set with 20 categories for China was collected based on a sampling strategy that combines automatic sample generation and visual interpretation. Simultaneously, a novel multi-stage method for fine-grained wetland classification was proposed, which integrates pixel-based and object-based strategies using ensemble learning algorithms and multi-source remote sensing data. First, a pixel-based ensemble learning algorithm was implemented to classify five rough wetland categories and six non-wetland categories. Second, an object-based ensemble learning approach was designed to separate the water cover in the pixel-based classification results into eight detailed categories. Third, the merged pixel-based and object-based classification results were refined with knowledge-based post-processing procedures to identify 14 fine-grained wetland categories. Results using the Pixel Information Expert Engine (PIE-Engine) cloud platform proved the effectiveness of the proposed wetland classification method. The overall accuracy, kappa, and weighted F1 reached 87.39%, 82.80%, and 86.02%, respectively. The adopted ensemble learning algorithm yielded better performance than classifiers such as CatBoost, random forest, and XGBoost. The incorporation of spectral, texture, shape, topographic, and geographic features from multi-source data contributed to differentiating wetland categories. According to the relative contribution, spectral indexes (NDVI and NDWI), texture features (sum average and contrast), and topographic features (slope and elevation) were identified as important leading predictors for the first-stage pixel-based classification. Shape features (shape index and compactness) and auxiliary features (geographic location) were crucial predictors for the second-stage object-based classification. Compared with other products, our 10-m wetland mapping results for national wetland reserves were rich in detail and fine in categories. Overall, the constructed sample set and developed classification method show promise in laying a foundation for large-scale wetland mapping. The derived wetland maps can provide support for wetland protection and restoration.

Mapping gully affected areas by using Sentinel 2 imagery and digital elevation model based on the Google Earth Engine

Gullies pose a great threat to the sustainable development of agricultural production and the ecological environment. Accurate mapping of gully affected areas plays an important role in regional environmental monitoring and management. In this paper, an object-based image analysis method based on the Google Earth Engine (OBIA-GEE) is proposed for mapping gully affected areas by using Sentinel 2 imagery and AW3D30 DEM data. The method utilizes the simple non-iterative clustering (SNIC) algorithm to segment the top-five principal component images derived from Sentinel-2 surface reflectance images. Then, the random forest (RF) algorithm is used for feature importance analysis and gully affected area mapping. Finally, the gully affected area mapping results are further optimized by establishing terrain skeleton lines. The proposed method is applied to five study areas with different landform types on the Chinese Loess Plateau, and the results show that the method achieves good performance in mapping gully affected areas, with an overall accuracy (OA) of 86.44%, an F score (F1) of 0.84, a user’s accuracy (UA) of 84.97%, and a producer’s accuracy (PA) of 83.90%. Compared with the OA of the results of a previous study (OA was 78.8%), the accuracy of the gully affected area mapping results herein improved by 7.64%. The RF feature importance results and findings of different feature combination cases demonstrate that textural information, time series spectral information, and topographic factors are important for mapping gully affected areas. Object-based and pixel-based classification results are compared, and the performance of the OBIA-GEE method is better than that of the pixel-based RF method. In addition, the OBIA-GEE method is highly efficient, which provides the possibility of applying the proposed method in large-scale studies. This research is beneficial to studies related to monitoring gullies and managing soil erosion.

Object-oriented polarimetric SAR image classification via the combination of a pixel-based classifier and a region growing technique

ABSTRACT Land-cover type interpretation by the use of remote sensing image classification techniques is always a hot topic. In this paper, an object-oriented method is presented for fully polarimetric synthetic aperture radar (SAR) image classification. Differing from most of the traditional object-oriented classification algorithms, the proposed method employs an innovative classification strategy that combines a pixel-based classifier and a region growing technique. Firstly, taking each individual pixel as a seed pixel, the homogeneous areas are extracted by a region growing technique. Then, using the information of the pixel-based classification result, the pixels located in each homogeneous area are all assigned to a certain class. Finally, the majority voting strategy is deployed to determine the final class label of each pixel. The experiments conducted on two fully polarimetric SAR images reveal that the proposed classification scheme can obtain pleasing classification accuracy and can provide the classification maps with more homogeneous regions than pixel-based classification.

Forest fire progress monitoring using dual-polarisation Synthetic Aperture Radar (SAR) images combined with multi-scale segmentation and unsupervised classification

Background The cloud-penetrating and fog-penetrating capability of Synthetic Aperture Radar (SAR) give it the potential for application in forest fire progress monitoring; however, the low extraction accuracy and significant salt-and-pepper noise in SAR remote sensing mapping of the burned area are problems. Aims This paper provides a method for accurately extracting the burned area based on fully exploiting the changes in multiple different dimensional feature parameters of dual-polarised SAR images before and after a fire. Methods This paper describes forest fire progress monitoring using dual-polarisation SAR images combined with multi-scale segmentation and unsupervised classification. We first constructed polarisation feature and texture feature datasets using multi-scene Sentinel-1 images. A multi-scale segmentation algorithm was then used to generate objects to suppress the salt-and-pepper noise, followed by an unsupervised classification method to extract the burned area. Key results The accuracy of burned area extraction in this paper is 91.67%, an improvement of 33.70% compared to the pixel-based classification results. Conclusions Compared with the pixel-based method, our method effectively suppresses the salt-and-pepper noise and improves the SAR burned area extraction accuracy. Implications The fire monitoring method using SAR images provides a reference for extracting the burned area under continuous cloud or smoke cover.

A High-Precision Crop Classification Method Based on Time-Series UAV Images

Timely and accurate information on crop planting structures is crucial for ensuring national food security and formulating economic policies. This study presents a method for high-precision crop classification using time-series UAV (unmanned aerial vehicle) images. Before constructing the time-series UAV images, Euclidian distance (ED) was utilized to calculate the separability of samples under various vegetation indices. Second, co-occurrence measures and the gray-level co-occurrence matrix (GLCM) were employed to derive texture characteristics, and the spectral and texture features of the crops were successfully fused. Finally, random forest (RF) and other algorithms were utilized to classify crops, and the confusion matrix was applied to assess the accuracy. The experimental results indicate the following: (1) Time-series UAV remote sensing images considerably increased the accuracy of crop classification. Compared to a single-period image, the overall accuracy and kappa coefficient increased by 26.65% and 0.3496, respectively. (2) The object-oriented classification method was better suited for the precise classification of crops. The overall accuracy and kappa coefficient increased by 3.13% and 0.0419, respectively, as compared to the pixel-based classification results. (3) RF obtained the highest overall accuracy and kappa coefficient in both pixel-based and object-oriented crop classification. RF’s producer accuracy and user accuracy for cotton, spring wheat, cocozelle, and corn in the study area were both more than 92%. These results provide a reference for crop area statistics and agricultural precision management.

Individual Tree Classification Using Airborne LiDAR and Hyperspectral Data in a Natural Mixed Forest of Northeast China

This paper proposes a method to classify individual tree species groups based on individual tree segmentation and crown-level spectrum extraction (“crown-based ITC” for abbr.) in a natural mixed forest of Northeast China, and compares with the pixel-based classification and segment summarization results (“pixel-based ITC” for abbr.). Tree species is a basic factor in forest management, and it is traditionally identified by field survey. This paper aims to explore the potential of individual tree classification in a natural, needle-leaved and broadleaved mixed forest. First, individual trees were isolated, and the spectra of individual trees were then extracted. The support vector machine (SVM) and spectrum angle mapper (SAM) classifiers were applied to classify the trees species. The pixel-based classification results from hyperspectral data and LiDAR derived individual tree isolation were compared. The results showed that the crown-based ITC classified broadleaved trees better than pixel-based ITC, while the classes distribution of the crown-based ITC was closer to the survey data. This indicated that crown-based ITC performed better than pixel-based ITC. Crown-based ITC efficiently identified the classes of the dominant and sub-dominant species. Regardless of whether SVM or SAM was used, the identification consistency relative to the field observations for the class of the dominant species was greater than 90%. In contrast, the consistencies of the classes of the sub-dominant species were approximately 60%, and the overall consistency of both the SVM and SAM was greater than 70%.

A Metric for Evaluating the Geometric Quality of Land Cover Maps Generated with Contextual Features from High-Dimensional Satellite Image Time Series without Dense Reference Data

Land cover maps are a key resource for many studies in Earth Observation, and thanks to the high temporal, spatial, and spectral resolutions of systems like Sentinel-2, maps with a wide variety of land cover classes can now be automatically produced over vast areas. However, certain context-dependent classes, such as urban areas, remain challenging to classify correctly with pixel-based methods. Including contextual information into the classification can either be done at the feature level with texture descriptors or object-based approaches, or in the classification model itself, as is done in Convolutional Neural Networks. This improves recognition rates of these classes, but sometimes deteriorates the fine-resolution geometry of the output map, particularly in sharp corners and in fine elements such as rivers and roads. However, the quality of the geometry is difficult to assess in the absence of dense training data, which is usually the case in land cover mapping, especially over wide areas. This work presents a framework for measuring the geometric precision of a classification map, in order to provide deeper insight into the consequences of the use of various contextual features, when dense validation data is not available. This quantitative metric, named the Pixel Based Corner Match (PBCM), is based on corner detection and corner matching between a pixel-based classification result, and a contextual classification result. The selected case study is the classification of Sentinel-2 multi-spectral image time series, with a rich nomenclature containing context-dependent classes. To demonstrate the added value of the proposed metric, three spatial support shapes (window, object, superpixel) are compared according to their ability to improve the classification performance on this challenging problem, while paying attention to the geometric precision of the result. The results show that superpixels are the best candidate for the local statistics features, as they modestly improve the classification accuracy, while preserving the geometric elements in the image. Furthermore, the density of edges in a sliding window provides a significant boost in accuracy, and maintains a high geometric precision.

Efficient Identification of Corn Cultivation Area with Multitemporal Synthetic Aperture Radar and Optical Images in the Google Earth Engine Cloud Platform

The distribution of corn cultivation areas is crucial for ensuring food security, eradicating hunger, adjusting crop structures, and managing water resources. The emergence of high-resolution images, such as Sentinel-1 and Sentinel-2, enables the identification of corn at the field scale, and these images can be applied on a large scale with the support of cloud computing technology. Hebei Province is the major production area of corn in China, and faces serious groundwater overexploitation due to irrigation. Corn was mapped using multitemporal synthetic aperture radar (SAR) and optical images in the Google Earth Engine (GEE) cloud platform. A total of 1712 scenes of Sentinel-2 data and 206 scenes of Sentinel-1 data acquired from June to October 2017 were processed to composite image metrics as input to a random forest (RF) classifier. To avoid speckle noise in the classification results, the pixel-based classification result was integrated with the object segmentation boundary completed in eCognition software to generate an object-based corn map according to crop intensity. The results indicated that the approach using multitemporal SAR and optical images in the GEE cloud platform is reliable for corn mapping. The corn map had a high F1-Score of 90.08% and overall accuracy of 89.89% according to the test dataset, which was not involved in model training. The corn area estimated from optical and SAR images was well correlated with the census data, with an R2 = 0.91 and a root mean square error (RMSE) of 470.90 km2. The results of the corn map are expected to provide detailed information for optimizing crop structure and water management, which are critical issues in this region.

Classification of Polarimetric SAR Data Based on Object-Based Multiple Classifiers for Urban Land-Cover

In this study, we used an object-oriented method for merging pixel-based classification and image segments to get an optimal classification result for an urban land-cover classification which is one of the important applications of Polarimetric SAR (PolSAR) remote-sensing. Because of the nature of PolSAR images, various features can be extracted and used in a classification. To achieve an improved classification accuracy, an optimal subset of features should be used. For this purpose, we used a class-based multiple classifier with a support vector machine (SVM) as a pixel-based classification with the class accuracy as the criterion in feature selection. We proposed the SVM margin as a new distance criterion for feature selection. In addition, to overcome the degradation of pixel-based classification results due to presence of speckle noise in PolSAR images, thematic features were used in image segmentation. In general, the proposed method consisted of three steps: feature selection, pixel-based classification, and polarimetric spatial classification. The pixel-based classification result was merged with a set of segments that were obtained from multi-resolution segmentation and the overall accuracy of the results was evaluated. The objectives of the study were to improve the accuracy of classification and to introduce margin as the new criterion for feature selection. Our results showed that the overall accuracy of the proposed method (90.07 %) was improved compared with the single SVM classifier (80.46 %) and pixel-based multiple classifier (83.61 %). Lastly, the SVM margin distance as well as class accuracy could be considered as appropriate criteria in feature selection.

A new combination classification of pixel- and object-based methods

This study presents a novel classification method using a combination of pixel- and object-based classifications, which includes the pixel-based uncertainty classification, object-based classification, and combined pixel- and object-based classification. As the spatial resolution increases, the complexity of land covers and the degree of information uncertainty in remote-sensing imagery increase, making remote-sensing image classification more difficult. For high-resolution image classification, using the pixel-based method it is easy to misclassify the different components with characteristic variations within the same land cover as different categories, and with the object-based method it is easy to misclassify the different categories of land cover with a strong spatial correlation as the same category. By using the proposed method, the pixel- and object-based classifications are performed on the image respectively, and the pixel-based classification result is utilized to correct the object-based classification result to obtain the optimized synthesis classification result. The experiments indicate that the combined classification method not only makes full use of the advantages of the individual-level methods, but also overcomes their disadvantages and produces higher classification accuracy than the single pixel- or object-based method. The accuracy improvement with the combined classification in the three experiments is 8.3, 9.5, and 13.2% relative to the pixel-based classification, and 7.2, 6.1, and 8.1% relative to the object-based classification.

Discriminating Saltcedar (Tamarix ramosissima) from Sparsely Distributed Cottonwood (Populus euphratica) Using a Summer Season Satellite Image

Accurate mapping of saltcedar (Tamarix ramosissima) and cottonwood (Populus euphratica) using remote sensing images is required to study the dynamic relationship between these two species. Our study used pixel-based and semi-object-based methods to classify a high spatial resolution QuickBird image acquired during the summer in northern China, where both saltcedar and cottonwood are native species. The pixel-based classification results revealed that spectral bands alone were not sufficient to discriminate saltcedar from cottonwood trees due to their similar foliage reflectance in the summer. Including texture measures did not improve the result. The unique crown shapes and shadows associated with sparsely distributed cottonwood were used to facilitate the semi object-based method. The overall accuracy of the object-based classification result increased 15 percent compared to that of the pixel-based classification results and showed significant improvement in the discrimination between saltcedar and cottonwood.

Monitoring urban land cover using satellite remote sensing techniques and field spectroradiometric measurements: case study of “Yialias” catchment area in Cyprus

This study highlights the need for digital mapping of sprawl phenomenon in catchment areas with the use of both satellite and ground remote-sensing techniques. The Yialias River, located in the central part of Cyprus, was selected as a case study area. In this catchment, devastating flash flood events took place in both 2003 and 2009, with catastrophic results. Initially, ground spectroradiometric measurements were applied to investigate the spectral similarity of different classes such as those of urban fabric and marl/chalk origin formations within the watershed. Temporal land cover changes were analyzed by using multitemporal satellite imagery (ASTER) and by incorporating both pixel- and object-oriented classification techniques. To create effective land use and land cover maps, a classification model was proposed based on spectral, texture, and shape characteristics. The pixel-based classification results were compared and evaluated with the object-based classification products. The optimum classification products were imported to geographical information systems and FRAGSTATS software and were used to visually and statistically detect landscape identifying trends based on spatial landscape metrics. The final results indicated considerable expansion within the study area during the last 10 years.

Object-based urban detailed land cover classification with high spatial resolution IKONOS imagery

Improvement in remote sensing techniques in spatial/spectral resolution strengthens their applicability for urban environmental study. Unfortunately, high spatial resolution imagery also increases internal variability in land cover units and can cause a ‘salt-and-pepper’ effect, resulting in decreased accuracy using pixel-based classification results. Region-based classification techniques, using an image object (IO) rather than a pixel as a classification unit, appear to hold promise as a method for overcoming this problem. Using IKONOS high spatial resolution imagery, we examined whether the IO technique could significantly improve classification accuracy compared to the pixel-based method when applied to urban land cover mapping in Tampa Bay, FL, USA. We further compared the performance of an artificial neural network (ANN) and a minimum distance classifier (MDC) in urban detailed land cover classification and evaluated whether the classification accuracy was affected by the number of extracted IO features. Our analysis methods included IKONOS image data calibration, data fusion with the pansharpening (PS) process, Hue–Intensity–Saturation (HIS) transferred indices and textural feature extraction, and feature selection using a stepwise discriminant analysis (SDA). The classification results were evaluated with visually interpreted data from high-resolution (0.3 m) digital aerial photographs. Our results indicate a statistically significant difference in classification accuracy between pixel- and object-based techniques; ANN outperforms MDC as an object-based classifier; and the use of more features (27 vs. 9 features) increases the IO classification accuracy, although the increase is statistically significant for the MDC but not for the ANN.

Object-oriented methods for habitat mapping at multiple scales – Case studies from Northern Germany and Wye Downs, UK

This paper presents an application of object-oriented techniques for habitat classification based on remotely sensed images and ancillary data. The study reports the results of habitat mapping at multiple scales using Earth Observation (EO) data at various spatial resolutions and multi temporal acquisition dates. We investigate the role of object texture and context in classification as well as the value of integrating knowledge from ancillary data sources. Habitat maps were produced at regional and local scales in two case studies; Schleswig-Holstein, Germany and Wye Downs, United Kingdom. At the regional scale, the main task was the development of a consistent object-oriented classification scheme that is transferable to satellite images for other years. This is demonstrated for a time series of Landsat TM/ETM+ scenes. At the local scale, investigations focus on the development of appropriate object-oriented rule networks for the detailed mapping of habitats, e.g. dry grasslands and wetlands using very high resolution satellite and airborne scanner images. The results are evaluated using statistical accuracy assessment and visual comparison with traditional field-based habitat maps. Whereas the application of traditional pixel-based classification result in a pixelised (salt and pepper) representation of land cover, the object-based classification technique result in solid habitat objects allowing easy integration into a vector-GIS for further analysis. The level of detail obtained at the local scale is comparable to that achieved by visual interpretation of aerial photographs or field-based mapping and also retains spatially explicit, fine scale information such as scrub encroachment or ecotone patterns within habitats.