Vol. 44 · No. 2

In mobile edge computing (MEC), the mobile device can offload tasks to the server at the edge of the mobile network for execution, thereby reducing the delay of task execution and the energy consumption of the mobile device. However, the limited resources of the edge server prevent the mobile device from offloading all tasks to the edge servers. To solve the problems, this paper constructs a multi-users and single edge server model for mobile edge computing. In order to minimize the weighted total cost combined energy consumption of mobile devices and task execution delay under the constraints of computing resources and storage resource of the edge server, we propose a task offloading decision and resource allocation algorithm based on improved deep deterministic policy gradient (DDPG) – PERDDPG. In our algorithm, a special reward function is designed to get the reward value for correlating negatively with the total cost. We can obtain the lowest total cost when the algorithm reaches the maximum reward value. Furthermore, we apply prioritized experience replay (PER) to improve DDPG. So, the PERDDPG has a more dynamic MEC scenario for making offloading decisions and computing resource allocation. Simulation results show that the proposed algorithm can get a better convergence speed and improve the cumulative reward compared to the existing algorithms, effectively reducing the weighted total cost of mobile devices and improving the success rate of task execution.

Breast cancer, a critical global health concern, necessitates accurate and timely diagnosis. This research introduces a novel methodology that harnesses modern technologies, including deep learning and transfer learning, to enhance breast cancer cell segmentation. The study commences with meticulous dataset selection and preprocessing, followed by image segmentation using advanced techniques to differentiate between benign and malignant cells effectively. Two significant algorithms, Convolutional Neural Networks (CNN) and AlexNet, are employed, achieving remarkable classification accuracy of 94.5

The goal of the 3D action prediction task is to predict the action label corresponding to an incomplete 3D skeleton sequence. Existing studies are limited to the supervised framework. To eliminate the dependence of supervised learning on expensive labels, we propose a self-supervised learning method for 3D action prediction. We use three self-supervised tasks of action completeness perception, motion prediction, and global regularization to allow the network to learn the past and future information embedded in the sequence of unfinished actions, i.e., the action completeness that has occurred and the future motion trend, and to optimize the feature space learned by the model. Some models ignore the past and future information embedded in partial sequences, which is the key to action prediction by humans. Based on our self-supervised method, we design two modules, an action completeness perceptron, and a motion predictor, to complete missing information in partial inputs. And a novel network structure is proposed to fuse partial and complete prediction to achieve more reasonable action prediction. We have conducted extensive experiments on different datasets, and the results validate the effectiveness of our proposed method.

Algorithms and data structures are principal parts of computer science education. For many students, however, it is not easy to master them due to their diversity and inherent complexity. The application of algorithm visualizations is a widely adopted approach, which can help to mitigate this difficulty. Within this work, we aim to improve the efficiency of the learning process in the field of algorithms and data structures. The main directions we use in this work to reach this goal are the introduction of comparative algorithm visualization and the implementation of the visualization tool based on contemporary standards. We analyze and compare several of the available solutions for algorithms and data structure visualization and evaluate them according to the provided functionalities. Further, we define a list of requirements, including the capability to compare selected algorithms visually. The practical outcome of this work is a web application that allows us to visualize and compare different algorithms and data structures in terms of their operation and efficiency. At the end of the paper, the proposed solution is evaluated in several ways.

Breast cancer, per WHO, ranks top in diagnoses and cancer fatalities. Early detection via mammography reduces mortality significantly, yet mammogram images often have indistinct features. Hence, precise tumor edge identification requires both image enhancement and segmentation. In response, we introduce the Max Planck Algorithm, a novel segmentation method rooted in Planck's quantum theory, specifically his thermal radiation principles. We innovatively converted this theory to create a unique segmentation tool applicable to digital image processing and medical imaging. The algorithm works by relating mammogram pixel values to X-ray wavelengths, adapting Planck's Law to use 'temperature' as an arbitrary variable (originally tied to actual temperature in Planck's work). Gradually adjusting 'temperature' optimizes the mammogram image's meaningfulness. The Max Planck Algorithm boasts advantageous properties, delivering higher efficiency and superior segmentation results. This innovative model introduces new methods for enhancing and segmenting mammograms, establishing itself as a unique technique without comparison to existing methods.

The common detection of fasteners of automobile door panels is based on the method of template matching, which has the problems of low detection accuracy and poor real-time performance under the influence of different lighting and different placement positions. To improve the detection speed and accuracy of fasteners in complex scenes, a small object detection algorithm, YOLO-DTO (Detect Tiny Object), was proposed based on the YOLOv8 algorithm. Firstly, considering that the algorithm uses strided convolution to compress the input image prematurely, resulting in the loss of fine-grained information in the early stage of the image, which makes it difficult to recover the complete detail information in the subsequent feature fusion process, this paper modifies the convolution module in the early stage of the algorithm and introduces the SPD (SPace-to-Depth) module to reconstruct the early stage of the original algorithm. Secondly, a selective attention module is embedded in the Neck output position of the algorithm to enhance the algorithm's ability to pay attention to the context information of fasteners. Finally, to optimize the regression efficiency of the bounding box, the MPDIoU loss function replaced the CIoU loss function. Experimental results show that the average detection accuracy of the YOLO-DTO algorithm is 98.8

The recognition of human behaviors in videos is a critical domain within human activity analysis. However, the current architectures and mechanisms of human behavior recognition methods dominated by CNN, GCNs, and LSTM are unduly complex resulting in high computational complexity of the models. Furthermore, these methods often exhibit poor robustness when it comes to recognizing behaviors across different environmental conditions and video angles. To address these challenges, this paper introduces a lightweight human skeleton interaction behavior inference network based on a multi-layer perceptron. This network leverages human skeleton information and utilizes minimal prior knowledge to infer limb behavior encoding. To reduce computational complexity, videos are divided into smaller segments, serving as the minimum computation units. This approach integrates three essential types of information: independent global information about individual postures, local interaction information regarding different limb parts, and temporal distance information. These three types of information are coupled through LSTM, incorporating temporal changes into network for recognition and classification. In comparison to previous similar methods, our proposed method is more lightweight, exhibits stronger robustness against interference and enables behavior recognition across different environments and perspectives.

A construction of a new family of three erasure correcting linear block codes over GF(q) with characteristic two together with their syndrome decoding procedures are presented in this paper. The designed code distance of four was confirmed by demonstrating a decoding algorithm capable of correcting three erasures. The second confirmation was obtained from the weight spectra of selected codes, which were calculated using Krawtchouck polynomials derived from the weight spectra of their dual codes.

Combating illegal drone activities is an important task for national defense and security. How to spot drones quickly and accurately is the key. While there are many ways to detect drones, their reasoning is generally slow and complex. Therefore, in this work, we propose an improved and efficient UAV detection method YOLOv8s-C3AS based on YOLOv8s. There are three main improvements to this approach: First, we propose a new Coordinate Channel Spatial Attention Module (CCSM) and add it to the backbone of the model to enable better feature extraction. Secondly, in order to solve the scale inconsistency problem of YOLOv8s PANet, we propose a new adaptive fusion feature network (PANet-AF), which enables the model to fuse the features of the three scales better, which enables the model to better integrate features of different scales. Third, we use a more reasonable bounding box regression loss function SIoU, which improves the detection accuracy of the model without cost. Finally, we refined and made public the drone dataset and conducted a series of experiments combined with the PASCCOL VOC dataset. Our proposed approach achieves 77.2

Scene recognition, an AI technology based on deep learning, has been widely used in public safety, road traffic, and automatic driving, but applying it on massive data in deep convolutional neural networks (DCNNs) results in performance bottlenecks. This paper proposes SMRFC-PDCNN, an efficient scene matching recognition algorithm that addresses three specific problems: decreased accuracy of feature maps, redundant feature calculations, and low efficiency in parallel recognition. The proposed algorithm includes a feature pooling selection strategy called MI-IPSS, a feature selection strategy called DCPSO-FSS, a load balancing strategy called CCG-LBS. MI-IPSS solves the problem of de-creased accuracy of feature maps by adapting the pooling strategy based on mutual information coefficient between feature maps before and after pooling. DCPSO-FSS uses density clustering and particle swarm optimization to locate clustering parameters quickly and recognize clustered features through sampling in the fully connected layer. CCG-LBS dynamically calculates the computing overhead of feature maps and allocates data between groups according to the over-head to solve the problem of low efficiency in parallel recognition. Experimental results show that SMRFC-PDCNN has good performance and is suitable for the fast scene matching recognition process of parallelized DCNNs on large-scale datasets.