Anomaly detection is a unique type of classification challenge. The coupling of imbalance, overlap and other complexity of the data such as noise in industrial internet of things (IIoT) scenarios affect the detection accuracy seriously. To address this issue, this article proposes a novel oversampling approach based on synthetic minority oversampling technology via hypersphere coverage and adaptive differential evolution (HCADE-SMOTE). However, overlap intensification caused by generated samples and over-loss of valid information have always been crucial problems for traditional SMOTE-based approaches. In HCADE-SMOTE, we identify error-prone samples including minority noise and boundary samples based on hypersphere coverage algorithm first. Then, we fine-tune the distribution of these error-prone samples before generation with an adaptive differential evolution algorithm. Instead of deletion mechanism, it avoids transitional information loss. With error-prone samples far away from the boundary, HCADE-SMOTE improves the boundary distribution and simplifies the judge the decision boundary for the detection models. Furthermore, minority samples are oversampled based on local hypersphere density and compactness with a weighted SMOTE mechanism to address imbalance problem. The superiority of this HCADE-SMOTE is verified by experiments from optimization of sample distribution, effect of anomaly detection, and statistical tests, compared with 7 well-known SMOTE-based methods. The experimental results show that HCADE-SMOTE is the most prominent to alleviate overlap with Fishers discriminant ratio metric. After HCADE-SMOTE, the detection results reached the best with classification metrics, for the four detection models Support Vector Machines (SVM), Logistic Regression (LR), Naive Bayes, and Decision Tree (DT). The statistical tests also prove HCADE-SMOTE has significant difference from other SMOTE-based methods, superior to them.

Three-dimensional (3-D) reconstruction of multi-view synthetic aperture radar (SAR) can effectively correct geometric distortion and recover the target scene intuitively. However, due to the drastic changes in SAR images during the multi-view observation, it would appear many false estimation points in the traditional 3-D reconstruction algorithms. This paper proposes a new 3-D reconstruction algorithm based on semantic information of ship structure to improve the 3-D reconstruction results with multi-view SAR. The semantic information is used to construct the 3-D solution space of the ship target, which can greatly improve the reconstruction accuracy. Simulations verify the effectiveness of the proposed algorithm.

Vision-language models, pre-trained on web-scale datasets, have the potential to greatly enhance the intelligence of web applications (e.g., search engines, chatbots, and art tools). Precisely, these models align disparate domains into a co-embedding space, achieving impressive zero-shot performance on multi-modal tasks (e.g., image-text retrieval, VQA). However, existing methods often rely on well-prepared data that less frequently contain noise and variability encountered in real-world scenarios, leading to severe performance drops in handling out-of-distribution (OOD) samples. This work first comprehensively analyzes the performance drop between in-distribution (ID) and OOD retrieval. Based on empirical observations, we introduce a novel approach, Evidential Language-Image Posterior (ELIP), to achieve robust alignment between web images and semantic knowledge across various OOD cases by leveraging evidential uncertainties. The proposed ELIP can be seamlessly integrated into general image-text contrastive learning frameworks, providing an efficient fine-tuning approach without exacerbating the need for additional data. To validate the effectiveness of ELIP, we systematically design a series of OOD cases (e.g., image distortion, spelling errors, and a combination of both) on two benchmark datasets to mimic noisy data in real-world web applications. Our experimental results demonstrate that ELIP improves the performance and robustness of mainstream pre-trained vision-language models facing OOD samples in image-text retrieval tasks.

Medium-Earth-orbit (MEO) synthetic aperture radar (SAR) could provide a wide observable area and short revisit times while offering a moderate spatial resolution. The combination of maneuvering platform SAR receivers and MEO SAR transmitters has become a new research hotspot. However, the receiver often works in nonlinear trajectory with high maneuverability and highly squinted angle in practice. The curved trajectory of MEO SAR and the complex motion of the receiver would cause severe azimuth variance and range-Doppler coupling. The existing imaging algorithms mostly suffer from azimuth defocusing if applied to MEO/high maneuvering platform receiver bistatic SAR (MEO/HMP-BiSAR). To solve these problems, this article proposes a modified extended azimuth nonlinear chirp scaling (EANLCS) algorithm with a new perturbation function. A high-precision echo model is deduced for the modified EANLCS algorithm to describe the range history and echo phase. The modified EANLCS algorithm not only solves the problem of range cell migration and severe range-Doppler coupling, but it also equalizes the spatial variance of Doppler frequency modulation (FM) rates as well as high-order Doppler coefficients in MEO/HMP-BiSAR. The proposed algorithm markedly surpasses conventional algorithms in image quality. Simulations and real experiments are conducted to verify the effectiveness of the proposed algorithm.

Automated graph learning has drawn widespread research attention due to its great potential to reduce human efforts when dealing with graph data, among which hyperparameter optimization (HPO) is one of the mainstream directions and has made promising progress. However, how to obtain reliable and trustworthy prediction results with automated graph neural networks (GNN) is still quite underexplored. To this end, we investigate automated GNN calibration by marrying uncertainty estimation to the HPO problem. Specifically, we propose a hyperparameter uncertainty-induced graph convolutional network (HyperU-GCN) with a bilevel formulation, where the upper-level problem explicitly reasons uncertainties by developing a probabilistic hypernetworks through a variational Bayesian lens, while the lower-level problem learns how the GCN weights respond to a hyperparameter distribution. By squeezing model uncertainty into the hyperparameter space, the proposed HyperU-GCN could achieve calibrated predictions in a similar way to Bayesian model averaging over hyperparameters. Extensive experimental results on six public datasets were provided in terms of node classification accuracy and expected calibration error (ECE), demonstrating the effectiveness of our approach compared with several state-of-the-art uncertainty-aware and calibrated GCN methods.