Applied Intelligence最新文献

In the field of computer vision, the task of human action recognition (HAR) represents a challenge, due to the complexity of capturing nuanced human movements from video data. To address this issue, researchers have developed various algorithms. In this study, a novel two-stream architecture is developed that combines LSTM with a depthwise separable convolutional neural network (DSConV) and skeleton information, with the aim of enhancing the accuracy of HAR. The 3D coordinates of each joint in the skeleton are extracted using the Mediapipe library, and the 2D coordinates are obtained using MoveNet. The proposed method comprises two streams, called the temporal LSTM module and the joint-motion module, and was developed to overcome the limitations of prior two-stream RNN models, such as the vanishing gradient problem and the difficulty of effectively extracting temporal-spatial information. A performance evaluation on the benchmark datasets of JHMDB (73.31%), Florence-3D Action (97.67%), SBU Interaction (95.2%), and Penn Action (94.0%) showcases the effectiveness of the proposed model. A comparison with state-of-the-art methods demonstrates the superior performance of the approach on these datasets. This study contributes to advancing the field of HAR, with potential applications in surveillance and robotics.

Graphical abstract

The initial segmentation phase is crucial in image processing to simplify the image representation and extract some desired features. Different methods and techniques have been proposed for image multi-level thresholding, but they are still stuck in local optima and need improvement. Recently, a metaheuristic optimization algorithm called Golden Jackal Optimizer (GJO) has been proposed as an alternative solution. The GJO has been adopted as a good solution for many optimization problems. However, the GJO attempted to solve the convergence problem to a local minimum during execution, often leading to unsatisfactory results. Most variants of GJO are based on chaotic systems due to their easy implementation and remarkable capacity to avoid being trapped in local optima. This paper proposes a Polynomial Chebychev Symmetric Chaotic-based GJO (PCSCGJO) algorithm by combining a recently developed chaotic generating function to achieve better segmentation results. This variant improves the GJO by introducing the chaotic generating function of the Chebyshev polynomials as an update process while searching for the optimal solution. Simulation results prove the effectiveness of the PCSCGJO method and its ability to deal with different medical color images. The quality of the segmented images obtained by the proposed method was compared to well-known metaheuristic algorithms using performance metrics such as PSNR, SSIM, FSIM, and MSE. Consequently, the metrics values show that the suggested technique outperforms the other methods regarding quality and accuracy.

Cross-Domain Few-Shot Learning (CDFSL) is one of the most cutting-edge fields in machine learning. It not only addresses the traditional few-shot problem but also allows for different distributions between base classes and novel classes. However, most current CDFSL models only focus on the generalization performance of high-level features during training and testing, which hinders their ability to generalize well to domains with significant gaps. To overcome this problem, we propose a CDFSL method based on Task Augmentation and Multi-Level Adaptive features representation(TA-MLA). At the feature representation level, we introduce a meta-learning strategy for multi-level features and adaptive features. The former come from different layers of network. They jointly participate in image prediction to fully explore transferable features suitable for cross-domain scenarios. The latter is based on a feature adaptation module of feed-forward attention, aiming to learn domain-adaptive features to improve the generalization of the model. At the training task level, we employ a plug-and-play Task Augmentation(TA) module to generate challenging tasks with adaptive inductive biases, thereby expanding the distribution of the source domain and further bridging domain gaps. Extensive experiments conducted on multiple datasets. The results demonstrate that our method based on meta-learning can effectively improves few-shot classification performance, especially in cases with significant domain shift.

Colorectal cancer is considered one of the deadliest diseases, contributing to an alarming increase in annual deaths worldwide, with colorectal polyps recognized as precursors to this malignancy. Early and accurate detection of these polyps is crucial for reducing the mortality rate of colorectal cancer. However, the manual detection of polyps is a time-consuming process and requires the expertise of trained medical professionals. Moreover, it often misses polyps due to their varied size, color, and texture. Computer-aided diagnosis systems offer potential improvements, but they often struggle with precision in complex visual environments. This study presents an enhanced deep learning approach using encoder-decoder architecture for colorectal polyp segmentation to capture and utilize complex feature representations. Our approach introduces an enhanced dual attention mechanism, combining spatial and channel-wise attention to focus precisely on critical features. Channel-wise attention, implemented via an optimized Squeeze-and-Excitation (S&E) block, allows the network to capture comprehensive contextual information and interrelationships among different channels, ensuring a more refined feature selection process. The experimental results showed that the proposed model achieved a mean Intersection over Union (IoU) of 0.9054 and 0.9277, a dice coefficient of 0.9006 and 0.9128, a precision of 0.8985 and 0.9517, a recall of 0.9190 and 0.9094, and an accuracy of 0.9806 and 0.9907 on the Kvasir-SEG and CVC-ClinicDB datasets, respectively. Moreover, the proposed model outperforms the existing state-of-the-art resulting in improved patient outcomes with the potential to enhance the early detection of colorectal polyps.

In the existing three-way conflict analysis models, ratings have only one scale. However, when evaluating an issue in real life, agents can also use multi-scale ratings, which can provide a more comprehensive description and better analysis. Therefore, it is necessary to study three-way conflict analysis based on a multi-scale situation table. In this paper, we consider the construction of three-way conflict analysis models on multi-scale situation tables (MS-STs). Firstly, we introduce the concept of MS-STs, in which the attitudes of agents towards issues are represented by multi-scale ratings. Secondly, we construct two types of three-way conflict analysis models on MS-STs using two different methods. One approach is to directly construct a three-way conflict analysis model on original MS-STs, called Type-1 three-way conflict analysis model. In this approach, we measure the conflict distances between agents on a subset of issues by using the proposed weighted distance function. We then trisect all pairs of agents. The other method involves converting an original MS-ST into a single-scale situation table through optimal scale selection. This results in a single-scale situation table induced by the optimal scale combination. Based on this, we construct a corresponding Type-2 three-way conflict analysis model. We provide several examples to illustrate the construction process of these two models. Additionally, we provide the calculation methods for weights and thresholds. Finally, we compare the proposed models in this paper with existing models to verify their applicability and effectiveness.

High-utility itemset mining (HUIM) is a widely adopted data mining technique for discovering valuable patterns in transactional databases. Although HUIM can provide useful knowledge in various types of data, it can be challenging to interpret the results when many patterns are found. To alleviate this, closed high-utility itemset mining (CHUIM) has been suggested, which provides users with a more concise and meaningful set of solutions. However, CHUIM is a computationally demanding task, and current approaches can require prolonged runtimes. This paper aims to solve this problem and proposes a meta-heuristic model based on particle swarm optimization (PSO) to discover CHUIs, called CHUI-PSO. Moreover, the algorithm incorporates several new strategies to reduce the computational cost associated with similar existing techniques. First, we introduce Extended TWU pruning (ETP), which aims to decrease the number of possible candidates to improve the discovery of solutions in large search spaces. Second, we propose two new utility upper bounds, used to estimate itemset utilities and bypass expensive candidate evaluations. Finally, to increase population diversity and prevent redundant computations, we suggest a structure called ExploredSet to maintain and utilize the evaluated candidates. Extensive experimental results show that CHUI-PSO outperforms the current state-of-the-art algorithms regarding execution time, accuracy, and convergence.

Cyberbullying has emerged as a pressing concern across various social platforms due to the escalating usage of online networks. Cyberbullying may lead victims to depression, self-harm, and even suicide. In this research, a two-stage cyberbullying detection framework based on multi-view features and decision fusion strategies is proposed. The first stage is to discover cyberbullying texts in social media, and the second stage delves into categorizing the specific forms of bullying present in the identified texts. In the two-stage detection process, features are constructed from multiple views, including Content view, Profanity view, and User view, to portray the bullying behavior. Furthermore, a decision fusion strategy is designed, incorporating both single-view features and multi-view features to enhance detection effectiveness. Finally, the research explains the complex mechanism of multi-view features in two-stage cyberbullying detection by calculating their SHAP values. The experimental results demonstrate the effectiveness of the multi-view feature and decision fusion strategy in cyberbullying detection. Notably, this framework yields impressive results, boasting an F1-score of 89.66% and an AUC of 95.98% in Stage I, while achieving an F1-score of 74.25% and an Accuracy of 79.01% in Stage II. The interpretability analysis of features affirms the pivotal role played by multi-view features, with the Content view features emerging as especially significant in the pursuit of effective cyberbullying detection.

Existing approaches rely on heuristic or learnable object proposals (which are required to be optimised during training) for 3D object detection. In our approach, we replace the hand-crafted or learnable object proposals with randomly generated object proposals by formulating a new paradigm to employ a diffusion model to detect 3D objects from a set of randomly generated and supervised learning-based object proposals in an autonomous driving application. We propose DDet3D, a diffusion-based 3D object detection framework that formulates 3D object detection as a generative task over the 3D bounding box coordinates in 3D space. To our knowledge, this work is the first to formulate the 3D object detection with denoising diffusion model and to establish that 3D randomly generated and supervised learning-based proposals (different from empirical anchors or learnt queries) are also potential object candidates for 3D object detection. During training, the 3D random noisy boxes are employed from the 3D ground truth boxes by progressively adding Gaussian noise, and the DDet3D network is trained to reverse the diffusion process. During the inference stage, the DDet3D network is able to iteratively refine the 3D randomly generated and supervised learning-based noisy boxes to predict 3D bounding boxes conditioned on the LiDAR Bird’s Eye View (BEV) features. The advantage of DDet3D is that it allows to decouple training and inference stages, thus enabling the use of a larger number of proposal boxes or sampling steps during inference to improve accuracy. We conduct extensive experiments and analysis on the nuScenes and KITTI datasets. DDet3D achieves competitive performance compared to well-designed 3D object detectors. Our work serves as a strong baseline to explore and employ more efficient diffusion models for 3D perception tasks.

The uncertainty of the stock market is the foundation for investors to obtain returns. Driven by interests, stock price forecasting has become a research hotspot. However, as the high latitude, highly volatile, and noisy, forecasting the stock prices has become a highly challenging task. The existing stock price forecasting methods only study low latitude data, which is unable to reflect the cumulative effect of multiple factors on stock price. To effectively address the high latitude, high volatility, and noise of stock price, a time-series information system (TSIS) forecasting approach for stock price is proposed. Aiming at dynamically depicting the real-world decision-making scenarios from a finer granularity, the TSIS is constructed based on the information systems. Then, a denoised neighborhood rough set (DNRS) model based on the TSIS is proposed by local density factor to achieve the purpose of feature selection, which can weaken the impact of noise on sample data. Subsequently, the multivariate empirical mode decomposition (MEMD) and multivariate kernel extreme learning machine (MKELM) are employed to decompose and forecast. Finally, the proposed TSIS forecasting approach is applied to stock price. Experimental results show that the TSIS forecasting approach for stock price has excellent performance and can be provided in the quantitative trading of stock market.

Due to the rapid economic development of modern society, the demand for cargo in the shipping industry has experienced unprecedented growth in recent years. The introduction of a large number of ships, especially large, new, and intelligent ships, has made shipping networks more complex. Controlling transportation risks has become more challenging than ever before. Ship trajectory prediction based on automatic identification system (AIS) data can effectively help identify abnormal ship behaviors and reduce maritime risks such as collisions, grounding, and contacts. In recent years, with the rapid development of deep learning theories, recurrent neural network models (long short-term memory and gated recurrent unit) have been widely used in ship trajectory prediction due to their powerful ability to capture hidden information in time-series data. However, these models struggle with tasks involving high complexity of trajectory features. To address this issue, this paper introduces a switching-input mechanism based on LSTM, constructing a ship trajectory prediction model based on the SI-LSTM model. The switching-input mechanism enables the model to adjust its processing of important information according to dynamic changes in input data, effectively capturing local features of complex trajectories. The experimental section, which includes eight cases of complex trajectories, demonstrates the competitive generalization ability and prediction accuracy of SI-LSTM.