Pattern Analysis and Applications最新文献

Unsupervised anomaly detection holds significant importance in large-scale industrial manufacturing. Recent methods have capitalized on the benefits of employing a classifier pretrained on natural images to extract representative features from specific layers, which are subsequently processed using various techniques. Notably, memory bank-based methods, which have demonstrated exceptional accuracy, often incur a trade-off in terms of latency, posing a challenge in real-time industrial applications where prompt anomaly detection and response are crucial. Indeed, alternative approaches such as knowledge distillation and normalized flow have demonstrated promising performance in unsupervised anomaly detection while maintaining low latency. In this paper, we aim to revisit the concept of knowledge distillation in the context of unsupervised anomaly detection, emphasizing the significance of feature selection. By employing distinctive features and leveraging different models, we intend to highlight the importance of carefully selecting and utilizing relevant features specifically tailored for the task of anomaly detection. This article presents a novel approach for anomaly detection, which employs dual model knowledge distillation and incorporates various types of semantic information by leveraging high and low-level semantic information.

Among several local descriptors invented in literature, the local binary pattern (LBP) is the prolific one. Despite its advantages like low computational complexity and monotonic gray invariance property, there are various demerits are observed in LBP and these are limited spatial patch, high dimension feature, noisy thresholding function and un-affective in harsh illumination variations. To overcome these issues presented work introduces the novel local descriptor called as discriminative binary pattern (DBP). Precisely two descriptors are introduced under DBP so-called Radial orthogonal binary pattern (ROBP) and radial variance binary pattern (RVBP). In former proposed descriptor, for neighborhood comparison, the center pixel is replaced by mean of medians computed from [orthogonal pixels + center pixel] of two 3 × 3 pixel window, formed from radius S1 and S2 of the 5 × 5 image patch. In latter proposed descriptor, the radial variances generated from 8 pair of two pixels are utilized for comparison with their mean value. In case of the both proposed descriptors, the sub-region wise histograms are extracted and fused to develop the entire feature size. Further the feature length of ROBP and RVBP are merged to form the size of the DBP descriptor. The compression is conducted by principal component analysis (PCA) and Fishers linear discriminant analysis). For matching support vector machines is used. Experiments conducted on 8 benchmark datasets reveals the effectiveness of the proposed DBP as compared to the other state of art benchmark methods.

Cognitive disorders affect various cognitive functions that can have a substantial impact on individual’s daily life. Alzheimer’s disease (AD) is one of such well-known cognitive disorders. Early detection and treatment of cognitive diseases using artificial intelligence can help contain them. However, the complex spatial relationships and long-range dependencies found in medical imaging data present challenges in achieving the objective. Moreover, for a few years, the application of transformers in imaging has emerged as a promising area of research. A reason can be transformer’s impressive capabilities of tackling spatial relationships and long-range dependency challenges in two ways, i.e., (1) using their self-attention mechanism to generate comprehensive features, and (2) capture complex patterns by incorporating global context and long-range dependencies. In this work, a Bi-Vision Transformer (BiViT) architecture is proposed for classifying different stages of AD, and multiple types of cognitive disorders from 2-dimensional MRI imaging data. More specifically, the transformer is composed of two novel modules, namely Mutual Latent Fusion (MLF) and Parallel Coupled Encoding Strategy (PCES), for effective feature learning. Two different datasets have been used to evaluate the performance of proposed BiViT-based architecture. The first dataset contain several classes such as mild or moderate demented stages of the AD. The other dataset is composed of samples from patients with AD and different cognitive disorders such as mild, early, or moderate impairments. For comprehensive comparison, a multiple transfer learning algorithm and a deep autoencoder have been each trained on both datasets. The results show that the proposed BiViT-based model achieves an accuracy of 96.38% on the AD dataset. However, when applied to cognitive disease data, the accuracy slightly decreases below 96% which can be resulted due to smaller amount of data and imbalance in data distribution. Nevertheless, given the results, it can be hypothesized that the proposed algorithm can perform better if the imbalanced distribution and limited availability problems in data can be addressed.

Graphical abstract

In the realm of 3D image processing, accurately representing the geometric nuances of line curves is crucial. Building upon the foundation set by the slope chain code, which adeptly represents intricate two-dimensional curves using an array capturing the exterior angles at each vertex, this study introduces an innovative 3D encoding method tailored for polygonal curves. This 3D encoding employs parallel slope and torsion chains, ensuring invariance to common transformations like translations, rotations, and uniform scaling, while also demonstrating robustness against mirror imaging and variable starting points. A hallmark feature of this method is its ability to compute tortuosity, a descriptor of curve complexity or winding nature. By applying this technique to biomedical engineering, we delved into the flagellar beat patterns of human sperm. These insights underscore the versatility of our 3D encoding across diverse computer vision applications.

For channel and spatial feature map C×W×H in object detection task, its information fusion usually relies on attention mechanism, that is, all C channels and the entire space W×H are all compressed respectively via average/max pooling, and then their attention weight masks are obtained based on correlation calculation. This coarse-grained global operation ignores the differences among multiple channels and diverse spatial regions, resulting in inaccurate attention weights. In addition, how to mine the contextual information in the space W×H is also a challenge for object recognition and localization. To this end, we propose a Fine-Grained Dual Level Attention Mechanism joint Spacial Context Information Fusion module for object detection (FGDLAM&SCIF). It is a cascaded structure, firstly, we subdivide the feature space W×H into n (optimized as n = 4 in experiments) subspaces and construct a global adaptive pooling and one-dimensional convolution algorithm to effectively extract the feature channel weights on each subspace respectively. Secondly, the C feature channels are divided into n (n = 4) sub-channels, and then a multi-scale module is constructed in the feature space W×H to mine context information. Finally, row and column coding is used to fuse them orthogonally to obtain enhanced features. This module is embeddable, which can be transplanted into any object detection network, such as YOLOv4/v5, PPYOLOE, YOLOX and MobileNet, ResNet as well. Experiments are conducted on the MS COCO 2017 and Pascal VOC 2007 datasets to verify its effectiveness and good portability.

Time series data are sequences of values that are obtained by sampling a signal at a fixed frequency, and time series classification algorithms distinguish time series into different categories. Among many time series classification algorithms, subseries-based algorithms have received widespread attention because of their high accuracy and low computational complexity. However, subseries-based algorithms consider the similarity of subseries only by shape and ignore semantic similarity. Therefore, the purpose of this paper is to determine how to solve the problem that subseries-based time series classification algorithms ignore the semantic similarity between subseries. To address this issue, we introduce the deep graph kernel technique to capture the semantic similarity between subseries. To verify the performance of the method, we test the proposed algorithm on publicly available datasets from the UCR repository and the experimental results prove that the deep graph kernel has an important role in enhancing the accuracy of the algorithm and that the proposed algorithm performs quite well in terms of accuracy and has a considerable advantage over other representative algorithms.

This study aims to address the challenges of detecting smoking behavior among workers in chemical plant environments. Smoking behavior is difficult to discern in images, with the cigarette occupying only a small pixel area, compounded by the complex background of chemical plants. Traditional models struggle to accurately capture smoking features, leading to feature loss, reduced recognition accuracy, and issues like false positives and missed detections. To overcome these challenges, we have developed a smoking behavior recognition method based on the YOLOv8 model, named Smoking-YOLOv8. Our approach introduces an SD attention mechanism that focuses on the smoking areas within images. By aggregating information from different positions through weighted averaging, it effectively manages long-distance dependencies and suppresses irrelevant background noise, thereby enhancing detection performance. Furthermore, we utilize Wise-IoU as the regression loss for bounding boxes, along with a rational gradient distribution strategy that prioritizes samples of average quality to improve the model’s precision in localization. Finally, the introduction of SPPCSPC and PConv modules in the neck section of the network allows for multi-faceted feature extraction from images, reducing redundant computation and memory access, and effectively extracting spatial features to balance computational load and optimize network architecture. Experimental results on a custom dataset of smoking behavior in chemical plants show that our model outperforms the standard YOLOv8 model in mean Average Precision (mAP@0.5) by 6.18%, surpassing other mainstream models in overall performance.

Molecular property prediction, crucial for early drug candidate screening and optimization, has seen advancements with deep learning-based methods. While deep learning-based methods have advanced considerably, they often fall short in fully leveraging 3D spatial information. Specifically, current molecular encoding techniques tend to inadequately extract spatial information, leading to ambiguous representations where a single one might represent multiple distinct molecules. Moreover, existing molecular modeling methods focus predominantly on the most stable 3D conformations, neglecting other viable conformations present in reality. To address these issues, we propose 3D-Mol, a novel approach designed for more accurate spatial structure representation. It deconstructs molecules into three hierarchical graphs to better extract geometric information. Additionally, 3D-Mol leverages contrastive learning for pretraining on 20 million unlabeled data, treating their conformations with identical topological structures as weighted positive pairs and contrasting ones as negatives, based on the similarity of their 3D conformation descriptors and fingerprints. We compare 3D-Mol with various state-of-the-art baselines on 7 benchmarks and demonstrate our outstanding performance.

We present a new recognition framework for ancient coins struck from the same die. It is called Low-complexity Arrays of Patch Signatures. To overcome the problem of illumination conditions we use multi-light energy maps which are a light-independent, 2.5D representation of the coin. The coin recognition is based on a local texture analysis of the energy maps. Descriptors of patches, tailored to coin images via the properties provided by the energy map, are matched against a database using a system of associative arrays. The system of associative arrays used for the matching is a generalization of the Low-complexity Arrays of Contour Signatures. Hence, the matching is very efficient and nearly at constant time. Due to the lack of available data, we present two new data sets of artificial and real ancient coins respectively. Theoretical insights for the framework are discussed and various experiments demonstrate the promising efficiency of our method.

Medical image datasets are essential for training models used in computer-aided diagnosis, treatment planning, and medical research. However, some challenges are associated with these datasets, including variability in data distribution, data scarcity, and transfer learning issues when using models pre-trained from generic images. This work studies the effect of these challenges at the intra- and inter-domain level in few-shot learning scenarios with severe data imbalance. For this, we propose a methodology based on Siamese neural networks in which a series of techniques are integrated to mitigate the effects of data scarcity and distribution imbalance. Specifically, different initialization and data augmentation methods are analyzed, and four adaptations to Siamese networks of solutions to deal with imbalanced data are introduced, including data balancing and weighted loss, both separately and combined, and with a different balance of pairing ratios. Moreover, we also assess the inference process considering four classifiers, namely Histogram, kNN, SVM, and Random Forest. Evaluation is performed on three chest X-ray datasets with annotated cases of both positive and negative COVID-19 diagnoses. The accuracy of each technique proposed for the Siamese architecture is analyzed separately. The results are compared to those obtained using equivalent methods on a state-of-the-art CNN, achieving an average F1 improvement of up to 3.6%, and up to 5.6% of F1 for intra-domain cases. We conclude that the introduced techniques offer promising improvements over the baseline in almost all cases and that the technique selection may vary depending on the amount of data available and the level of imbalance.