Mobile QR Code QR CODE

Multi-objective Cluster-head-based Energy-aware Routing using a Separable Convolution Neural Network in a Wireless Sensor Network

(Danish Ather);(J. Prisca Mary);(Pooja Singh);(Kanika Garg);(T.R. Priyadharshini);(B. Anni Princy);(Mohit Tiwari)

Wireless sensor network (WSN) applications are added day by day owing to numerous global uses (by the military, for monitoring the atmosphere, in disaster relief, and so on). Here, trust management is a main challenge. Sensor nodes are important in wireless sensor networks, but they are easily depleted because of their short lifespan from continuous sensing activity and low battery capacity. So efficient energy utilization is a challenging task in a WSN. To minimize energy loss, clustering with an optimum path selection process is needed to retain energy in sensor nodes. This manuscript proposes multi-objective Pelican Optimization Algorithm (POA) routing to maintain energy efficiency and minimize transmission distances in wireless sensor networks. A cluster head (CH) is selected by using a Separable Convolution Neural Network (SCNN). Simulation outcomes prove the proposed technique attains 22.3% and 25.04% improvements in energy consumption when compared to the Multi-Objective CH Energy-aware Optimized Routing Approach at WSN (MOCH-EORA-WSN) and Multiple Optimum Cluster Head Multi-Objective Grasshopper Optimization with Harmony-search at WSN (MOCH-MOGOH-WSN), respectively.

Live Streaming Interaction and Content Information Dissemination on TikTok Short Video Platform

(Ziwei Zhao);(Chunying Ge)

Short video platforms also allow the sharing of short videos and live streaming. Users can inde-pendently search for live streaming content for distribution, but the efficiency is relatively low. For example, the TikTok platform usually employs intelligent recommendation algorithms to actively distribute live streaming content to users in order to enhance the dissemination efficiency of live streaming content. This paper introduces a model of a user's live streaming interaction behavior on short video platforms, which was integrated with a collaborative filtering algorithm to build a rec-ommendation algorithm. A case analysis was carried out using the TikTok short video platform. The combined algorithm was compared with an algorithm based on association rules and one based on collaborative filtering. It was found that the combined recommendation algorithm per-formed the best when the number of nearest neighbor users was 15. Compared with the other two recommendation algorithms, the combined recommendation algorithm has a higher recom-mendation accuracy and more consistent recommendations of network anchor types.

Label-efficient Learning Methods for Computer Vision Applications

(Moon Ye-Bin);(Tae-Hyun Oh)

In this work, we review the challenges of data scarcity and label inefficiency in deep learning and survey efforts to overcome these challenges. Many label-efficient learning methods have been proposed, but there is still room to develop more effective methods. We introduce potential yet promising directions to achieve label-efficient learning in terms of data, learning methods, and efficient use of prior knowledge. We also present case studies involving the latest methods.

Modeling Method for Classification of Piano Music Style based on Big Data Mining and Machine Learning

(Wei You)

With the progress of music digitalization, various styles of music have been produced, and effective classification of music has become an important research direction. In this research, a model for piano-music style classification was constructed based on big data mining and machine learning algorithms. The input music signal was dealt with using framing, signal enhancement, and windowing. The Meldor Frequency Coefficient (MFC) and emotional features in the signal were extracted and fused to obtain combined features. The extracted feature vectors were input into a Deep Belief Network (DBN) for training and then a hidden Markov model (HMM) for classification and recognition. However, it was found that during the HMM training process, the algorithm produces large differences in the randomly selected initial matrix parameters, which cause the results to be trapped at a local optimum and affect the accuracy of model classification and recognition. To optimize the parameters, a genetic algorithm was used to optimize the classification model. The average Relative Percent Difference (PRD) was 2.402, the run time was 2.117 s, and the accuracy was 97.074%, which means the model can efficiently and accurately classify piano music styles.

R-CNN Auto-system for Detecting Text Road Signs in Baghdad

(Omar M. S. Ali);(Ali A. D. Al-Zuky);(Fatin E. M. Al-Obaidi)

Due to inadequate lighting, motion blur, occlusion, and the eventual disappearance of road signs, the determination of textual road signs is difficult to resolve. With the aid of a recurrent convolutional neural network (R-CNN), the current study focuses on detecting textual road signs in Baghdad at different times of day under varied situations, including vehicle speed, surrounding layers, epochs of the R-CNN, etc. Two types of different contrast on signs were used: blue and blue-green signs with white text. The differences in contrast seem to play an effective role in recall, sensitivity, and F1 score values. Results showed that the precision values for all signs and epochs were unity. For 20 and 60 epochs, the sensitivity values for the blue sign were 47.43% and 48.35%, respectively, while for the blue-green sign, the sensitivity values were equal to 95.19% for both numbers of epochs. The F1 scores were 0.6435 and 0.9753 for 20 epochs, while for 60 epochs it was 0.6518 and 0.9753 for blue and blue-green signs, respectively. The experiments validated the suggested software and provided implementation guidance to diagnose and automatically classify text road signs on streets.

Aerobics Exercise Performance Training for College Students based on Optimized Apriori Algorithm

(Li Shang)

Amazing changes have occurred in college education with the rapid development and popularization of science and technology in the information age. This paper proposes a design method for predicting the effect of college students’ sports performance training based on the optimized Apriori algorithm to develop a more standardized and complete physical education course training plan. The research introduces the gravitational search algorithm (GSA), and the particle swarm optimization algorithm (PSA) combines the two algorithms into a hybrid algorithm that integrates with the Apriori algorithm to form the GSA-PSO-A algorithm to predict the performance training effect of students. The algorithm was used to find valuable associated data in the data and finally conduct application analysis. The GSA-PSO algorithm reached a stable fitness value when the number of iterations was 200 and 100 in the unimodal and multimodal function tests, respectively, and the feasibility was the best. The GSA-PSO-A algorithm proposed by the research institute could effectively mine the training data of college students’ sports performance and provide a feasible path for improving the teaching of physical education courses to college students.

Performance Improvement of a Virtual Network Embedding Algorithm based on Temporal-difference Learning by Resource-Constraint-Aware Candidate Solution Selection

(Yukinobu Fukushima);(Yuta Sagawa);(Yuya Tarutani);(Tokumi Yokohira)

In this paper, we tackle a virtual network embedding problem in network virtualization. For this problem, an algorithm (VNE-TD) based on temporal difference learning has been proposed. VNE-TD, however, does not consider the node and link resource constraints in selecting the candidate solutions. Therefore, when attempting to construct a virtual network, the embedding of a virtual network may fail due to insufficient resources. In this paper, we modify VNE-TD to select only those candidate solutions that satisfy the node and link resource constraints. We add a function to check the satisfiability of the node and link resource constraints to VNE-TD. The simulation results show that our modified methods reduce the blocking ratio of virtual network requests by up to 80% compared to VNE-TD.

Innovative Thinking for AI-based Ceramic Art Design

(Baoqing Zhang)

In the history of art, ceramics has held an irreplaceable position from ancient times to the present. Present day ceramic art design is monotonous and lacks personalization and innovation, which not only makes it difficult to meet the esthetic and usage needs of the public for ceramic art, but also hinders its sustainable development. To address this dilemma, this article discusses the use of artificial intelligence (AI) combined with innovative thinking in ceramic art design. Based on demand analysis of ceramic art design, AI interaction technology converts design variables into integrable output and input signals. Then, interaction between designers and design variables is achieved through model connection. The relevant design variables and data are classified and extracted as a whole, and a set of element resources is generated. Finally, innovative ceramic art design is achieved through user demand analysis. To verify the effectiveness of innovative thinking in ceramic art design based on AI, practical evaluations are conducted from four aspects: visual image, product value, functional utility, and innovation. The practical results show that the average scores for content innovation, form innovation, and concept innovation in ceramic art design based on AI were 8.04, 8.16, and 8.23 points out of 10, respectively. The practical results indicate that AI-based innovative thinking about ceramic art design has a certain operability and important value in promoting innovative development in ceramic art.

Wideband Speech Codec Algorithm based on Compressed Sensing and Fractional Calculus

(Xiuhuan Wang)

Broadband speech encoding and decoding methods are important for achieving high-quality speech communication and audio applications. However, encoding and decoding algorithms often face problems such as large data-transmission volume and high computational complexity. To overcome these problems, a wideband speech codec algorithm is proposed based on compressed sensing and fractional calculus. Compressed sensing theory was used to sparsely represent wideband speech signals. The concept and method of fractional calculus are introduced to analyze and process wideband speech signals. Algebraic codebooks were used to adapt the structure and bit allocation of speech based on its different states and actual encoding and decoding rates. Embedded encoding and decoding of wideband speech can be achieved by adding and generating digital book pulses layer by layer. The results show that the proposed algorithm has a minimum encoding and decoding rate of 6.9 bits/s and a speech quality score of over 4.0. It also has low latency and high speed for speech encoding and decoding and provides high-quality speech evaluation. It has clear advantages in speech quality and data transmission efficiency. This study could provide new ideas and methods for further research and application of broadband speech coding and decoding.

Contraband Identification Algorithm for Intelligent Millimeter-wave Security Screening Device based on Regional-convolution Neural Network Algorithm for Civil Aviation

(Fengwei Chen);(Tianyou Wu)

Security is directly tied to the protection of public property and lives and is one of the key safeguards for the regular functioning of civil aviation. Millimeter-wave-based security screening technology has been developed to handle the demand for security screening during periods of high passenger traffic and to minimize the involvement of security personnel. But it cannot meet the need for quick passage at peak passenger flow because the present millimeter-wave contraband image content is insufficient, and the target-detection accuracy is low. In order to solve this problem, this study examined the denoising of millimeter-wave contraband images based on mean filtering and the wavelet transform, and the Canny algorithm was used to realize the edge detection of images. A mask region-convolutional neural network algorithm was used to identify contraband targets in the detection area to realize the real-time monitoring of millimeter-wave security equipment. The peak signal-to-noise ratio and mean square error of the mean filtered denoising-wavelet transform algorithm were 25.439 dB and 65.4781, respectively. The classification accuracy rates were greater than those of the fast area convolutional neural network model-based approach (93.65%, 89.94%, and 91.25%, respectively). In conclusion, the suggested algorithm is reliable and effective at locating and identifying targets for contraband in civil aviation security screening.