Guanyuan Shuai

Edge-pixels-based support vector data description for specific land-cover distribution mapping

Abstract. An edge-pixels-based support vector data description (EPSVDD) method has been developed for improving one-class classification accuracy. The proposed method was validated in two experiments: a simulated experiment and an actual experiment. In the simulated experiment, a ring segmentation search method was performed to segment the wheat spectral feature for deriving training samples of different spectral responses. As the training data moved from the center to the edge of wheat distribution, the hypersphere expanded and the overall accuracy (OA) simultaneously increased, highlighting the potential advantage of edge pixels in SVDD classification. In the actual experiment, edge training samples were manually acquired from geographical parcel boundary and minimum noise fraction (MNF) scatterplots for both wheat and bare-land classes. For the wheat class, EPSVDD yielded an improved classification with an OA of 92.71% and a producer’s accuracy of 95.81%, which were higher than those of conventional SVDD method using typical training samples. Similarly, for the bare-land class, the OA of the EPSVDD was 92.53%, which was also significantly higher than traditional SVDD method. Then, SVDD classifications were carried out and repeated 10 times using different training set sizes. Mean OAs were almost higher than 0.9 with variance less than 0.03 using edge training samples, while highest OAs for wheat and bare land classes were 0.74 and 0.81, respectively, using random sampling method. The EPSVDD can effectively select the informative training sample for SVDD classifier to improve the accuracy of one-class classification.

Improving sampling efficiency of crop acreage estimation using wheat planting rule from historical remote sensing

Abstract The technology of remote sensing combining sampling is an effective way to estimate crop acreage (CA) at large scale. Previous research proved that if the crop proportion within a sampling unit is sufficiently stable from year to year, pixels classified from historical remote sensing images could offer reliable regression estimators for current CA. However, previous works explored various approaches for CA estimation using one-year historical data, which makes it difficult to determine which year has the highest correlation to the targeted year, especially for a region where background information about the cultivated planting system is scarce. We estimated the winter wheat acreage of Beijing in 2009 by using two stratification variables including the coefficient of variance and the mean CA of the sampling units via a two-stage stratified sampling method with multiple historical remotely sensed data. Results show that: (1) our method has higher sampling average accuracy and lower standard errors of sample averages than simple random sampling or the one-stage stratification method with CA as the auxiliary variable, which are the usual methods employed in the previous studies. (2) Fewer samples are required to get the predefined accuracy, which reduces costs. (3) Generally, using mean CA derived from multiyear historical remotely sensed data as the auxiliary variable has a higher accuracy than those data using CA derived from one-year historical remote sensing images as the auxiliary variable in one-stage sampling.

Support vector data description model to map urban extent from National Polar-Orbiting Partnership Satellite–Visible Infrared Imaging Radiometer Suite nightlights and normalized difference vegetation index

Abstract. We explored a one-class classifier, the support vector data description (SVDD), using the Suomi National Polar-Orbiting Partnership Satellite–Visible Infrared Imaging Radiometer Suite and normalized difference vegetation index to map the urban extent, which was tested in the Beijing and Tianjin city group area. The urban edge-pixels were selected as training samples for SVDD based on a profile-based sampling method combining nighttime light value histograms. The results showed that the overall accuracy of SVDD was similar to the support vector machine (SVM) model. However, kappa coefficients of SVDD for highly developed cities were superior to SVM, as producer and user accuracies of SVDD were almost equal to show high agreement of urban and nonurban areas. For metropolitan areas, such as Beijing and Tianjin, the urban extent generated by SVDD is closer to the reference data. The R2 between the quantity of SVDD-estimated urban extent and population, 0.86, was higher than that obtained from SVM, 0.76, indicating that the estimated urban extent from the SVDD is more efficient for understanding the population development. The SVDD was further applied for three other representative metropolitans in China: Shanghai, Guangzhou, and Shenzhen to validate the SVDD’s performance, and similar results were achieved. The success of the SVDD-based urban extent extraction improves our ability to map urban extent at regional and national scales.

Dryland summer crop classification using multi-temporal RADARSAT-2 images and objects information from optical image

This paper has proposed a new method that integrates the advantage of optical image for delineating land surface boundaries and the superiority of PolSAR data for obtaining corn information despite bad weather conditions. The comparison between the proposed method and both pixel- and object-based method was made to test their performance for corn classification. The analysis shows that the proposed method can significantly improve the overall accuracy and kappa value of the other two methods, which indicates that the proposed method was suitable for crop mapping using optical and Radarsat-2 PolSAR data.

KFDA-based cropland inundation change detection with an automatic method for training sample extraction

Flood is the most frequent disaster in the world, which can do harm to agriculture and threat to food security. Using kernel based supervised classifier to execute change detection for multi-temporal remote sensing data is a common method for flood disaster monitoring and assessment, and kernel Fishers discrimination analysis (KFDA) is one of them. Choosing training sample by visual interpretation is an important step, but difficult and wasting time, for the reason that a great amount of the flooded pixels are heterogeneous. In this study, we proposed an automatic sample extraction method, finding pixels in relative homogeneous areas by multiresolution segmentation and zonal standard deviation calculating, and then assigning sample class via clustering or linear discrimination of some specific index. The results showed that overall accuracy could reach 91.57% and the Kappa coefficient was 0.8316. The method we proposed was proved to be efficient.

SVDD-based one-class land-cover mapping using optimal training samples

Remotely sensed data have been widely used in the field of producing land-cover thematic maps. When dealing with single class problem, one-class classifiers proved to be more effective compared with conventional supervised classifiers. The Support Vector Data Description (SVDD), one kind of one-class classification method, has been applied to specific land-cover classifications lately. However, the sampling scheme used in previous studies does not follow the SVDD principle. In this paper, Euclidean distance and Mahalanobis distance were chosen as an index to optimize training samples in order to improve the accuracy of SVDD classification. Result shows that sample optimization do improve the classification accuracy. Besides, compared with the Euclidean distance, Mahalanobis distance is more suitable and effective for sample optimization.

Crop distribution mapping using hard and soft change detection method with multi-temporal remote sensing images

To take advantage of conventional hard land use/cover change detection method (HLUCD) and soft land use/cover change detection method (SLUCD), we develop a soft and hard land use/cover change detection method (SHLUCD) for crop distribution mapping. Two HJ-1/CCD images, acquired on 6 October 2011 (T1) and 16 April 2012 (T2) which represented the period of sowing and jointing respectively, were utilized by SHLUCD to extract wheat area in study area. The results show that the crop distribution derived from the SHLUCD reflects reality more accurately than that from HLUCD and SLUCD. Crops distribution mapping derived from SHLUCD give lowest RMSE and bias and the highest R2 than that from other two methods in all window size. Wheat distribution in typical area and mixed pixels zone could be identified by land use/cover change status and land change scope respectively through SHLUCD. Moreover, the theory and methods employed in developing the SHLUCD provide a new way for crop distribution mapping based on change detection technique.

SVDD-based land-cover mapping using optimal parameters via single window flexible pace search method

The SVDD method, one of the most popular one-class classifiers, could use training samples of the interest class to derive accurate classification and thus is adopted in this paper. However, the penalty parameter C and the kernel width s should be tuned carefully to construct an optimal hypersphere. This research developed a single window flexible pace search method to select optimal parameters. First, 120 edge pixels were acquired from parcel boundary and PCA image. Then a 3*3 window was applied to the training samples to obtain the buffer training set. Then optimal parameters were select through the flexible pace search method. Under optimal parameters, the buffer training set yielded an accurate classification with an overall accuracy of 89.70%, which differed slightly with that derived from the SVM classification. Thus, we conclude that our proposed method could be used to select optimal parameters for the SVDD method.