Concurrency and Computation-Practice & Experience最新文献

Manual measurement of cattle body size presents challenges, such as inducing stress responses in animals and inefficiencies. For large livestock like cattle, measurement based on full point clouds involves extensive computations and interference between different cloud sections. To address this, we propose MMLG-Point, a novel deep learning model for cattle point cloud segmentation and body size measurement, which introduces a Multilevel Geometric Perception Encoder and a Transformer-based decoder architecture. The encoder integrates Kernel Point Convolution (KPConv) and Separable Structure-Aware Learning (SSAL) with residual multiscale fusion to capture local geometric structures of large livestock point clouds, while the decoder employs CrossNorm and SelfNorm (CNSN) modules to enhance generalization under limited labeled data. Furthermore, an unsupervised pretraining strategy based on masked point reconstruction is proposed, enabling the model to learn structural and semantic representations from unlabeled cattle point clouds. Experimental results demonstrate that MMLG-Point achieves outstanding segmentation accuracy with minimal supervision, obtaining an overall accuracy (OA) of 94.3% and a mean Intersection over Union (mIoU) of 89.4% on the Simmental cattle dataset using only 12 labeled samples. The model also exhibits strong cross-species generalization, achieving 92.3% OA and 86.7% mIoU on pig datasets. Based on segmentation results, an automatic cattle body measurement algorithm is developed, incorporating density analysis, curvature detection, and contour extraction to compute parameters such as withers height, hip height, body length, chest girth, and abdominal circumference, achieving a mean absolute percentage error (MAPE) below 6%. These results confirm that the proposed MMLG-Point framework provides an effective and generalizable approach for high-precision segmentation and measurement of large livestock point clouds.

This paper presents a novel triple-color watermarking algorithm that leverages the elliptical monogenic wavelet transform (EMWT) and singular value decomposition (SVD) for robust and secure watermark embedding. The proposed method first encrypts the host image using a random matrix. Subsequently, both the host and watermark images undergo preprocessing techniques, including EMWT, discrete wavelet transform (DWT), discrete cosine transform (DCT), and SVD, to obtain the diagonal matrices required for embedding. For enhanced security, the watermark images are encrypted using the Arnold method with phase and linear representation embedding, where the second key is derived from the host image via Fourier transform. The watermarked image is then generated by applying the corresponding inverse transforms. Experimental results demonstrate the algorithm's excellent embedding and extraction performance on color images and icons with characters. Notably, the proposed method exhibits strong robustness against various attacks, including non-geometric attacks (e.g., noise, filtering), geometric attacks (e.g., rotation, cropping), and mixed attacks, outperforming existing color watermarking methods in terms of anti-attack ability and robustness.

Cloud computing is the massive evolution in Information Technology (IT) that provides virtualized and scalable resources to an end-user with minimum maintenance and cost. However, this environment is vulnerable to various attacks. These attacks impose heavy damages and affect the performance of the cloud. Therefore, timely detection of attacks in cloud computing is crucial. Hence, an efficient detection model is required for identifying various attacks in cloud computing. In this research, Shuffle Attention Convolutional Forward Fractional Network (SACFFNet) is presented to detect attacks in cloud computing. Initially, the cloud is simulated and next, a recorded log file is obtained from certain datasets. Thereafter, feature scaling is accomplished employing the minimum-maximum method. Then, features are selected using chord distance and Support Vector Machine Recursive Feature Elimination (SVM-RFE). Afterwards, data augmentation is done utilizing Synthetic Minority Oversampling Technique (SMOTE). Lastly, attacks are detected employing the newly designed SACFFNet. The SACFFNet is designed by integrating Shuffle Attention Network (SA-Net) with Convolutional Neural Network (CNN) with the modification of layers based on Fractional Calculus (FC). Additionally, SACFFNet has attained 91.678% of accuracy, 92.884% of sensitivity, and 92.090% of specificity.

Blockchain sharding has emerged as a promising solution to address the scalability challenges of modern blockchain systems, yet in practice, existing sharding systems still suffer from serious performance bottlenecks. In recent work, LB-Chain proposes a framework for load balancing by dynamically migrating hot accounts, which significantly improves throughput and latency. However, the approach still suffers from the computational overhead associated with frequent full ordering, which limits its efficiency in large-scale systems. To address this issue, this paper proposes an enhanced sharding framework, DH-Chain, which improves computational efficiency by introducing heap sorting optimization, dynamically managing the load of sharding and accounts, and avoiding full-volume sorting during each round of migration. The experimental results show that DH-Chain achieves 3.2% higher throughput than LB-Chain and 11.4% higher than random allocation, approaching the theoretical upper bound of ideal allocation. By leveraging heap-based sorting and two-phase commit protocols, DH-Chain ensures atomicity and security while reducing computational overhead. The framework effectively balances shard loads, maintaining consistent performance across varying transaction loads and demonstrating robust scalability.

Data mining is a technique that involves evaluating big data in order to reveal patterns or connections that can aid in solving business problems through the analysis of data. Exceptional model mining (EMM) is a system for finding local patterns in a dataset by identifying subgroups whose behavior deviates substantially from that which is anticipated. Incremental learning allows the model to continuously learn from new examples while retaining prior knowledge. In the instance of EMM, incremental learning increases the process by adding new training samples, which helps to enhance the model's performance over time. This research presents an incremental learning approach in EMMiner (InEMMiner) using a deep learning model. A novel hybrid optimizing algorithm, double exponential-running city game optimizer (DE-RCGO), is proposed by putting together double exponential smoothing (DES) and running city game optimizer (RCGO) for the purpose of finding outstanding subgroups in within-individual clusters. To assess quality of these subgroups, four different metrics are suggested: (1) a combination of weighted Kraskov entropy with a support vector machine (SVM), (2) a combination of weighted Fuzzy entropy with an Autoencoder, (3) a combination of weighted Boltzmann entropy with a gated recurrent unit (GRU), and (4) a metric that combines weighted Gibbs entropy with isolation forest. Such measures are incorporated in the EMMiner system in order to effectively filter and prioritize most appropriate exceptional subgroups with an improvement on overall EMM task performance. InEMMiner incorporates an incremental learning approach using a deep learning model, TabNet, which is further optimized by the proposed DE-RCGO algorithm. Experimental results validate the efficiency of the model, with quality metric achieving a value of 98.9%, memory usage of 985.65 MB, and a computational time of 81.52 s.

With the rapid development of software-defined networking (SDN), the single-controller architecture is unable to meet the performance and reliability requirements of the whole system. Consequently, a distributed multicontroller architecture has been proposed, in which the number and locations of controllers must be determined rigorously, formulating the controller placement problem (CPP). In order to solve the CPP by optimizing the propagation latency, we propose a convolutional node embedding and K-means algorithm (CNEKA), which integrates information propagation among adjacent nodes and calculation of embedding vectors by graph convolutional networks (GCNs) with the graph segmentation by K-means algorithm. As far as we know, we are the first paper to apply GCN to solving CPP. The study demonstrates that the CNEKA algorithm significantly enhances performance in optimizing average and worst-case latency between controllers and switches, as well as the propagation latency of the whole network, in varied experimental conditions. CNEKA achieves up to a 64.18% reduction in propagation latency when compared to other algorithms, and maintains a high stability with a fluctuation less than 7.6% when repeating the same experiments for several times. Moreover, CNEKA can always find the optimal or near-optimal solution with an error less than 11.15% when compared with the global optimal solution.

Hyperspectral image (HSI) classification demands high-performance processing and accuracy due to data dimensionality and computational complexity. The support vector machine (SVM) algorithm, which is renowned for its ability to manage high-dimensional data through kernel techniques for classification tasks effectively, has shown promising results in HSI analysis and applications. Composite kernels enhance SVM efficacy by integrating multiple kernels to capture diverse data characteristics. Existing SVM algorithms based on composite kernels are susceptible to high latency, computationally intensive operations, and poor scalability, which makes it challenging to deploy efficiently on resource-limited high-performance platforms like field-programmable gate arrays (FPGAs). The paper proposes a novel composite weighted-summation kernel (WSK) that combines polynomial and hardware friendly kernels (HFKs). The WSK kernel has been integrated with SVM, resulting in a classifier named WSK-SVM. Furthermore, a scalable FPGA-based system on chip (SoC) architecture is developed for HSI classification using WSK-SVM. The architecture implements a fast kernel approximation method to compute a HFK which eliminates multiplications. The architecture leverages spatial and temporal parallelism, efficiently aggregating processing into clusters of parallel binary WSK-SVM classifiers using the one-vs-one (OvO) methodology. The approach is scalable and achieves a level of spatial parallelism that facilitates the concurrent classification of multiple pixels. Two different SoC boards were used to implement the design. One board was based on the Xilinx Zynq-7000 and featured four parallel classifiers, and the second board was based on the Zynq UltraScale+ MPSoC and featured eleven parallel classifiers. High classification accuracy (>∖,99%) was achieved on benchmark datasets (Indian Pines, Pavia Centre, and Pavia University) using 8-bit fixed-point processing, resulting in negligible loss of floating-point accuracy. Resource utilization on the PYNQ-Z2 reached 78.37% for LUTs and 77.14% for BRAM, whereas on the ZCU-104, it was 49.76% for LUTs and 95.19% for BRAM. Real-time latency was as low as 1.26 $��\mu �s�$ per pixel (ZCU-104, Pavia Centre dataset) with a throughput of up to 6.35 MBPS. The proposed architecture provides scalable and high-performance solutions for onboard HSI classification under real-time constraints.