Low‐rank preserving embedding regression for robust image feature extraction

Abstract

Although low‐rank representation (LRR)‐based subspace learning has been widely applied for feature extraction in computer vision, how to enhance the discriminability of the low‐dimensional features extracted by LRR based subspace learning methods is still a problem that needs to be further investigated. Therefore, this paper proposes a novel low‐rank preserving embedding regression (LRPER) method by integrating LRR, linear regression, and projection learning into a unified framework. In LRPER, LRR can reveal the underlying structure information to strengthen the robustness of projection learning. The robust metric L2,1‐norm is employed to measure the low‐rank reconstruction error and regression loss for moulding the noise and occlusions. An embedding regression is proposed to make full use of the prior information for improving the discriminability of the learned projection. In addition, an alternative iteration algorithm is designed to optimise the proposed model, and the computational complexity of the optimisation algorithm is briefly analysed. The convergence of the optimisation algorithm is theoretically and numerically studied. At last, extensive experiments on four types of image datasets are carried out to demonstrate the effectiveness of LRPER, and the experimental results demonstrate that LRPER performs better than some state‐of‐the‐art feature extraction methods.

Deep graph contrastive learning model for drug-drug interaction prediction

“…Dimensionality reduction methods mainly include linear and nonlinear approaches [ 61 ]. Linear LRPER [ 62 ] and the nonlinear method T-SNE [ 63 ] are two popular dimensionality reduction methods. Given T-SNE’s advantage in preserving local structure, we opt for T-SNE as our dimensionality reduction tool.…”

“…Dimensionality reduction methods mainly include linear and nonlinear approaches [61]. Linear LRPER [62] and the nonlinear method T-SNE [63] In contrast, DrugBank performs optimally with only 1-Hop neighbors due to its higher node degree, where a larger subgraph might introduce noise, potentially counteracting the benefits. In the concluding model, the parameter H is set to 2 for the BioSNAP and AdverseDDI datasets, whereas it is set to 1 for the DrugBank dataset.…”

Relaxation linear regression with spectral–spatial constrained locality adaptive regularization for hyperspectral image classification

“…However, the computational cost of SVM is very high when the number of classes is relatively large. To treat this problem, linear regression (LR) was developed for classification [5, 6]. However, the LR‐based methods exactly transform the labels into a strict binary matrix, which results the label fitting with few freedom.…”

Leveraging Hyperspectral Images for Accurate Insect Classification with a Novel Two-Branch Self-Correlation Approach

“…Based on existing visual insect recognition research in recent years, with the development of visual insect recognition research, this study briefly divides these works into legal computer vision technology-based and deep learning-based methods. Traditional computer vision methods, exemplified by works such as [3,4], rely on feature extraction and automated classification, but they often exhibit lower accuracy or limited generalization. By contrast, applying deep learning based on convolutional neural networks (CNNs) to insect recognition can more accurately identify insects without the need for manual feature extraction.…”