The high-dimensional data space of hyperspectral images (HSIs) often result in ill-conditioned formulations, which finally leads to many of the high-dimensional feature spaces being empty and the ...useful data existing primarily in a subspace. To avoid these problems, we use distance metric learning for dimensionality reduction. The goal of distance metric learning is to incorporate abundant discriminative information by reducing the dimensionality of the data. Considering that global metric learning is not appropriate for all training samples, this paper proposes an ensemble discriminative local metric learning (EDLML) algorithm for HSI analysis. The EDLML algorithm learns robust local metrics from both the training samples and the relative neighborhood of them and considers the different local discriminative distance metrics by dealing with the data region by region. It aims to learn a subspace to keep all the samples in the same class are as near as possible, while those from different classes are separated. The learned local metrics are then used to build an ensemble metric. Experiments on a number of different hyperspectral data sets confirm the effectiveness of the proposed EDLML algorithm compared with that of the other dimension reduction methods.
Feature distortions of data are a typical problem in remote sensing image classification, especially in the area of transfer learning. In addition, many transfer learning-based methods only focus on ...spectral information and fail to utilize spatial information of remote sensing images. To tackle these problems, we propose spectral-spatial weighted kernel manifold embedded distribution alignment (SSWK-MEDA) for remote sensing image classification. The proposed method applies a novel spatial information filter to effectively use similarity between nearby sample pixels and avoid the influence of nonsample pixels. Then, a complex kernel combining spatial kernel and spectral kernel with different weights is constructed to adaptively balance the relative importance of spectral and spatial information of the remote sensing image. Finally, we utilize the geometric structure of features in manifold space to solve the problem of feature distortions of remote sensing data in transfer learning scenarios. SSWK-MEDA provides a novel approach for the combination of transfer learning and remote sensing image characteristics. Extensive experiments have demonstrated that the proposed method is more effective than several state-of-the-art methods.
Target Detection Based on Random Forest Metric Learning Dong, Yanni; Du, Bo; Zhang, Liangpei
IEEE journal of selected topics in applied earth observations and remote sensing,
04/2015, Letnik:
8, Številka:
4
Journal Article
Recenzirano
Target detection is aimed at detecting and identifying target pixels based on specific spectral signatures, and is of great interest in hyperspectral image (HSI) processing. Target detection can be ...considered as essentially a binary classification. Random forests have been effectively applied to the classification of HSI data. However, random forests need a huge amount of labeled data to achieve a good performance, which can be difficult to obtain in target detection. In this paper, we propose an efficient metric learning detector based on random forests, named the random forest metric learning (RFML) algorithm, which combines semimultiple metrics with random forests to better separate the desired targets and background. The experimental results demonstrate that the proposed method outperforms both the state-of-the-art target detection algorithms and the other classical metric learning methods.
Display omitted
•The maximum margin metric learning method (MMML) was applied to process geochemical exploration data.•The Mahalanobis distance was used to measure the similarity of samples.•A case ...study based on Fe–polymetallic mineralization in the Southwestern Fujian Province, China, is presented.
Geochemical anomaly identification is an important task in mineral exploration targeting. This task can be regarded as a binary classification problem whereby the aim is to discriminate between anomalous and not anomalous (i.e., background). We can analyze geochemical data from the aspects of frequency distributions, correlations and variances, geometrical properties of geochemical anomalies, and scale independence of geochemical patterns. In this study, we recognize geochemical anomalies based on the intrinsic relationship between geochemical elements which can be addressed by measuring their separation distance. As a machine learning method, metric learning aims at exploiting the statistical information between the geochemical features in limited training samples, and developing a more suitable distance to evaluate the similarity of samples without priori distribution assumptions. Accordingly, a metric learning method based on the maximum margin frame was applied to identify multivariate geochemical anomalies related to Fe-polymetallic mineralization in the southwestern Fujian Province of China. The geochemical exploration data were firstly translated into a dimensional reduced subspace with the help of maximum margin metric learning (MMML). The adaptive coherence estimator (ACE) detector, from the field of target detection, was then employed to identify geochemical anomalies. The anomaly results obtained by this integrated procedure using a combination of MMML and ACE were compared to those obtained by ACE without metric learning using a receiver operating characteristic (ROC) curve. The area under the curve value implied superior performance using the combination of MMML and ACE, suggesting that this hybrid method can be effectively applied to recognize geochemical anomalies linked to mineralization.
Rare metals play a considerable role in the development of new materials and energy, making them key mineral resources for global competition. Widely distributed along the Himalayan orogen, the ...Himalayan leucogranite belt is expected to be an important rare metal metallogenic belt in China. Thus, mapping the spatial distribution of Himalayan leucogranites is critical for prospecting rare metal deposits. The distribution characteristics of geochemical elements are important indicators for lithological identification. The differences in mineral composition and major oxide content between leucogranites and the surrounding rocks facilitate lithological mapping. However, significant uncertainty could arise owing to limited geochemical data due to particularly adverse working conditions and to difficulty in handling similar geochemical data. In this study, a metric learning-based approach is used for mapping leucogranites based on regional geochemical exploration datasets. Defined as a measure of similarity between two samples, metric learning reveals a “better distance” by converting original data into a more suitable Mahalanobis metric space with maximum separation of the target and the background. In this approach, a local weighted metric learning method is first used to assign weights to the training samples in the neighborhood, with respect to their reconstruction contributions in learning the local metric. Then, a discriminative local ensemble learning method is employed to integrate all learned metrics and to convert the original geochemical data into a metric space. This enables more effective separation of highly similar target leucogranites from the surrounding rocks with the help of a support vector machine. The distribution of leucogranites mapped by such a hybrid method showed high consistency with the geological map, indicating that this approach is reasonable for providing the indicated signature of leucogranites mapping in the study area. These results further provide an alternative way for identifying favorable intrusions based on geochemical exploration data.
•Himalayan leucogranites belt is expected to be an important rare metal metallogenic belt in China.•A hybrid of method of metric learning and support vector machine was used for mapping Himalayan leucogranites.•The obtained results have a close spatial relationship of mapped leucogranites.
By using the high spectral resolution, hyperspectral images (HSIs) provide significant information for target detection, which is of great interest in HSI processing. However, most classical target ...detection methods may only perform well based on certain assumptions. Simultaneously, using limited numbers of target samples and preserving the discriminative information is also a challenging problem in hyperspectral target detection. To overcome these shortcomings, this paper proposes a novel adaptive information-theoretic metric learning with local constraints (ITML-ALC) for hyperspectral target detection. The proposed method firstly uses the information-theoretic metric learning (ITML) method as the objective function for learning a Mahalanobis distance to separate similar and dissimilar point-pairs without certain assumptions, needing fewer adjusted parameters. Then, adaptively local constraints are applied to shrink the distances between samples of similar pairs and expand the distances between samples of dissimilar pairs. Finally, target detection decision can be made by considering both the threshold and the changes between the distances before and after metric learning. Experimental results demonstrate that the proposed method can obviously separate target samples from background ones and outperform both the state-of-the-art target detection algorithms and the other classical metric learning methods.
With the rapid development in the global economy and technology, urbanization has accelerated. It is important to characterize the urban expansion and determine its driving force. In this study, we ...used the Xiaonan District in Hubei Province, China, as an example to map and quantify the spatiotemporal dynamics of urban expansion from the two perspectives of built-up area and urban land in 1990–2020 by using remote sensing images. The location of rivers was found to be a primary limiting factor for spatial patterns and expansion of the built-up area. The transfer of the city center and the main direction of expansion generally corresponded well to the topography, policies, and development strategies. The built-up area expanded faster than the urban population in 1995–2020, which caused a waste in land resources. The results showed that the urban expansion first decreased and then increased during the research period. The increase in the proportion of the secondary industry was the main driving force of the urban expansion. Based on the characteristics of urban expansion in the past three decades, we conclude that the urban land of Xiaonan District will expand quickly in the future and will occupy vast agricultural land. The government must deploy control measures to balance the benefits and costs of urban expansion.
Recently, deep learning has developed rapidly, while it has also been quite successfully applied in the field of hyperspectral classification. Generally, training the parameters of a deep neural ...network to the best is the core step of a deep learning-based method, which usually requires a large number of labeled samples. However, in remote sensing analysis tasks, we only have limited labeled data because of the high cost of their collection. Therefore, in this paper, we propose a deep metric learning with online hard mining (DMLOHM) method for hyperspectral classification, which can maximize the inter-class distance and minimize the intra-class distance, utilizing a convolutional neural network (CNN) as an embedded network. First of all, we utilized the triplet network to learn better representations of raw data so that raw data were capable of having their dimensionality reduced. Afterward, an online hard mining method was used to mine the most valuable information from the limited hyperspectral data. To verify the performance of the proposed DMLOHM, we utilized three well-known hyperspectral datasets: Salinas Scene, Pavia University, and HyRANK for verification. Compared with CNN and DMLTN, the experimental results showed that the proposed method improved the classification accuracy from 0.13% to 4.03% with 85 labeled samples per class.
Hyperspectral change detection (CD) aims to obtain the change information of objects in the multitemporal hyperspectral images (HSIs). Recently, with the advantages in fully extracting the image ...features of irregular areas, the graph convolutional network (GCN) has attracted increasing attention for hyperspectral CD. The existing GCN-based CD methods usually use a graph structure constructed by superpixels to reduce the computational cost, which ignores the multiorder difference information among graph nodes and the local difference information within superpixels. To address these problems, this article proposes an efficient multiorder GCN with a channel attention module (CAM) for hyperspectral CD. Specifically, the multiorder GCN module is designed by repeatedly mixing the feature representations of neighborhoods. The CAM is then proposed to enhance the difference features of bitemporal HSIs. After that, the pixel-wise CD is accomplished by a lightweight feature fusion module and a fully connected layer. Experiments on three hyperspectral datasets illustrated the effectiveness of the proposed algorithm.
•GF-5 and Sentinel-2 fusion dataset is used for lithological mapping.•A new lithological classification method (ViT-DGCN) is proposed.•ViT-DGCN innovatively combines transformer and dynamic graph ...convolution.•ViT-DGCN gives superior performance compared with other methods.
Lithological identification and mapping using remote sensing (RS) imagery are challenging. Traditional lithological mapping relies mainly on multispectral data and machine learning methods. However, inadequate spectral information and inappropriate classification algorithms are major problems for RS geological applications. Moreover, satellite hyperspectral images (HSI) at low spatial resolution and convolutional neural network (CNN)-based methods with incomplete feature extraction remain challenging because of the limitations of sensor imaging and convolutional kernels for lithological mapping. To address the above issues, in this study, smoothing filter-based intensity modulation (SFIM) fusion technology is first employed to fuse GaoFen-5 hyperspectral images and Sentinel-2B multispectral images. This approach significantly improves spatial details and enriches spectral information. Subsequently, a novel Vision Transformer Dynamic Graph Convolutional Network (ViT-DGCN) is proposed for lithological mapping of the Cuonadong dome, Tibet, China. ViT-DGCN is a joint model consisting of a transformer and a dynamic graph convolution module that enhances feature extraction capabilities by exploring long-range interaction sequence features and dynamic graph structure information in a targeted manner. The proposed algorithm exhibits superior performance compared to the others, achieving an overall accuracy of 97% for the Cuonadong dome using only 1% of the available training samples.