Composite Kernel Method Research Articles

FPGA Acceleration of a Composite Kernel SVM for Hyperspectral Image Classification

Hyperspectral image classification is one of the most important techniques for analyzing hyperspectral image that have hundreds of spectrum luminance values of near-infrared to visible light. For this classification, supervised learning methods are widely used, but in general, they typically trade off between their accuracy and computational complexity. Our approach is based on a composite kernel method, and the computation is simplified to achieve higher processing speeds by efficiently using the hardware resources of FPGA. The accuracy of this approach reaches 98.0% and 98.8% on two benchmark datasets, Indian Pines and Salinas via simulation, which is comparable to those in previous works. Two implementations, one with less hardware resources but more off-chip memory bandwidth, and another with more hardware resources but less off-chip memory bandwidth, are implemented on an FPGA and evaluated. The processing speeds of the two implementations are the same, which is 1.3 Mpixels/ <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$s$ </tex-math></inline-formula> for 2048 pixel wide images. This processing speed is fast enough for real-time processing, and faster than previous studies when normalized by hardware size and power consumption. We also introduce two more implementations that aim to reduce the on-chip memory usage of the second implementation within a reasonable increase of off-chip memory bandwidth, and we discuss which implementation is advantageous under what conditions.

Multiscale Weighted Adjacent Superpixel-Based Composite Kernel for Hyperspectral Image Classification

This paper presents a composite kernel method (MWASCK) based on multiscale weighted adjacent superpixels (ASs) to classify hyperspectral image (HSI). The MWASCK adequately exploits spatial-spectral features of weighted adjacent superpixels to guarantee that more accurate spectral features can be extracted. Firstly, we use a superpixel segmentation algorithm to divide HSI into multiple superpixels. Secondly, the similarities between each target superpixel and its ASs are calculated to construct the spatial features. Finally, a weighted AS-based composite kernel (WASCK) method for HSI classification is proposed. In order to avoid seeking for the optimal superpixel scale and fuse the multiscale spatial features, the MWASCK method uses multiscale weighted superpixel neighbor information. Experiments from two real HSIs indicate that superior performance of the WASCK and MWASCK methods compared with some popular classification methods.

Composite reduced-kernel weighted extreme learning machine for imbalanced data classification

In order to solving the problem that the weighted extreme learning machine based on the ensemble learning method enhances the classification performance while increasing the running time of the algorithm, starting from the perspective of multi-core learning, a weighted extreme learning machine based on composite kernel functions and reduced-kernel technique is proposed. The composite kernel function based on Gaussian kernel and Polynomial kernel weighted combination is designed, which effectively improves the classification performance of weighted extreme learning machine. Meanwhile, based on the sample distribution characteristics of the imbalanced dataset, the balanced input sub-matrix is designed to reduce the computational cost of the composite kernel method. The eight binary classification imbalanced datasets of KEEL dataset repository were used for testing. The experimental results show that compared with the original weighted extreme learning machine algorithm, the G- mean and AUC classification performance indicators of the composite reduced-kernel weighted extreme learning machine algorithm are improved in each dataset, and the computation cost is effectively reduced.

Distribution feeder‐level day‐ahead peak load forecasting methods and comparative study

Day-ahead peak load forecasting is a crucial factor for the electrical utility for daily power planning and distribution. A comparative study of the peak load forecasting at distribution feeder-level circuits with high penetration level of PV generation are investigated. First, two years of private load data from the local utility and two years of public load data from Texas utility are analysed. The correlation analysis is applied to peak load and its driver factors. Next, Bayesian additive regression trees (BART) is employed to do peak load forecasting. Since the BART method takes an amount of time to generate the forecasting, the composite kernel methods based on Gaussian process regression (CKGPR) are designed. Then these methods are compared with multiple linear regression method and the support vector regression method based on the residential area and the business area load data. Thorough comparison results are presented based on five forecasting measurements. The BART has the best forecasting accuracy among all the indices, and the CKGPR also has counterpart forecasting results but with less computation time. Meanwhile, the forecasting accuracy difference between two areas is analysed. Lastly, influential driver factors are summarised.

Gabor feature‐based composite kernel method for hyperspectral image classification

Different from the traditional kernel classifiers that map the original data into high-dimensional kernel space, a novel classifier that projects Gabor features of the hyperspectral image into the kernel induced space through composite kernel technique is presented. The proposed method can not only improve the flexibility of the exploitation of spatial information but also successfully apply the kernel technique from a very different perspective to strengthen the discriminative ability. Experiments on the Indian Pines dataset demonstrate the superiority of the proposed method.

Hyperspectral image classification by AdaBoost weighted composite kernel extreme learning machines

Abstract Extreme learning machine (ELM) is an efficient learning algorithm for multi-classification and regression. However, original ELM doesn't consider the weight of each sample in training-set, which may cause the accuracy decreasing especially in imbalanced datasets. Even if each training sample is assigned with an extra weight, the problem on how to determinate the weight adaptively still remains. Inspiration by AdaBoost algorithm, we embed the weighted ELM algorithm in AdaBoost framework. In the meanwhile, we incorporate spatial and spectral information in composite kernel for each sample, which has a good performance in hyperspectral image (HSI) classification. By combining composite kernel methods and Adaboost framework with weighted ELM, a novel algorithm, namely AdaBoost composite kernel extreme learning machines denoted as AdaBoost-WCKELM is proposed. Experimental results demonstrate that the proposed method outperforms current state-of-the-art algorithms and derives a good improvement in HSI classification accuracy.

Composite Kernel Method for PolSAR Image Classification Based on Polarimetric-Spatial Information

The composite kernel feature fusion proposed in this paper attempts to solve the problem of classifying polarimetric synthetic aperture radar (PolSAR) images. Here, PolSAR images take into account both polarimetric and spatial information. Various polarimetric signatures are collected to form the polarimetric feature space, and the morphological profile (MP) is used for capturing spatial information and constructing the spatial feature space. The main idea is that the composite kernel method encodes diverse information within a new kernel matrix and tunes the contribution of different types of features. A support vector machine (SVM) is used as the classifier for PolSAR images. The proposed approach is tested on a Flevoland PolSAR data set and a San Francisco Bay data set, which are in fine quad-pol mode. For the Flevoland PolSAR data set, the overall accuracy and kappa coefficient of the proposed method, compared with the traditional method, increased from 95.7% to 96.1% and from 0.920 to 0.942, respectively. For the San Francisco Bay data set, the overall accuracy and kappa coefficient of the proposed method increased from 92.6% to 94.4% and from 0.879 to 0.909, respectively. Experimental results verify the benefits of using both polarimetric and spatial information via composite kernel feature fusion for the classification of PolSAR images.

A KERNEL METHOD BASED ON TOPIC MODEL FOR VERY HIGH SPATIAL RESOLUTION (VHSR) REMOTE SENSING IMAGE CLASSIFICATION

Abstract. A kernel-based method for very high spatial resolution remote sensing image classification is proposed in this article. The new kernel method is based on spectral-spatial information and structure information as well, which is acquired from topic model, Latent Dirichlet Allocation model. The final kernel function is defined as K = u1Kspec + u2Kspat + u3Kstru, in which Kspec, Kspat, Kstru are radial basis function (RBF) and u1 + u2 + u3 = 1. In the experiment, comparison with three other kernel methods, including the spectral-based, the spectral- and spatial-based and the spectral- and structure-based method, is provided for a panchromatic QuickBird image of a suburban area with a size of 900 × 900 pixels and spatial resolution of 0.6 m. The result shows that the overall accuracy of the spectral- and structure-based kernel method is 80 %, which is higher than the spectral-based kernel method, as well as the spectral- and spatial-based which accuracy respectively is 67 % and 74 %. What's more, the accuracy of the proposed composite kernel method that jointly uses the spectral, spatial, and structure information is highest among the four methods which is increased to 83 %. On the other hand, the result of the experiment also verifies the validity of the expression of structure information about the remote sensing image.