Vol. 42 · No. 6

Recent research shows that the effect of heterogeneous graph embedding learning is vulnerable to non-attribute nodes. However, the existing methods mainly use first-order neighbor nodes to complete attributes, which cannot achieve a satisfactory completion effect on the heterogeneous graphs with random non-attribute nodes. Therefore, this paper put forward an attribute completion method of heterogeneous graphs based on attribute neighborhoods, which is called HGCAN. HGCAN employs two major stages of completion. Specifically, in the first stage, we use meta-paths to construct attribute neighborhoods of non-attribute nodes. The attribute neighborhoods aggregation can capture the semantic relations of attributed nodes to initially complete attributes. Then, the second stage uses structural information to obtain the distance relationships between nodes to further improve the preliminary completed attributes. Finally, HGCAN is combined with an existing heterogeneous graph embedding learning model to verify the validity of the completed attributes and make the system end-to-end. Extensive experiments carried out on the ACM dataset show the proposed mechanism's superior performance over state-of-the-art attribute-completion methods.

Quantum-behaved Bat Algorithm (QBA) has been successfully applied as an optimal technique for dealing with a variety of optimization problems. Nevertheless, QBA suffers from similar problems as other swarm intelligent algorithms, such as poor exploration search and falling into local optima in certain conditions. Aiming at these shortcomings, an improved algorithm that combines convergence factor and gold sinusoidal self-learning mutation strategies (CGQBA) is proposed. A directional convergence factor is designed for the global position update process, it can improve the exploration search ability of the algorithm. Meanwhile, a self-learning predictive mutation mechanism is added to the algorithm. It contributes to the algorithm to jump out of the local extremum. The improved CGQBA algorithm is tested on 20 test functions with different characteristics in the numerical simulation experiments. The results and statistical tests show that CGQBA algorithm has better convergence speed, accuracy and stability. What is more, the multi-threshold image segmentation is modelled as an optimization problem, CGQBA algorithm is applied to the optimization problem to further verify the effectiveness and practicability in the real-world optimization. The results compared with three classical segmentation methods illustrate that CGQBA algorithm can effectively solve the image segmentation problem. It has a better segmentation effect and anti-noise ability.

Vehicle detection in intelligent transportation systems usually adopts cloud computing mode. The increasing amount of traffic surveillance video has brought challenges to the storage, communication, and processing of intelligent transportation systems based on cloud computing models. In this paper, we propose a vehicle detection scheme based on edge computing. First, the traffic surveillance video is preprocessed at the edge device. Using the frame difference algorithm based on structural similarity (SSIM) to remove video redundant frames, and avoid repeated frames in the subsequent extracted key frame sequence. Then, a frame difference algorithm based on local maxima is used to extract key frames as the basis for subsequent vehicle detection. Finally, the YOLOv5s is improved and used for vehicle detection. The efficient channel attention mechanism (ECA) is introduced to enhance the important features of the vehicle and suppress the general features to strengthen the detection network's ability to identify vehicle targets. At the same time, the Focal loss function is introduced to solve the positive and negative sample imbalance problem and improve the detection speed. The experimental results show that the scheme has more advantages than the original YOLOv5s in terms of precision, recall, and mean average precision.

Frequent Itemsets Mining (FIM), which consists of extracting frequent patterns from a transactional database, is considered one of the most successful techniques in data mining. Generally, the FIM problem can be solved by either the exact or metaheuristic-based methods. Exact methods, such as the Apriori algorithm, are highly effective for dealing with small to medium datasets. However, these methods need more temporal complexity when dealing with large datasets. Metaheuristic-based methods are becoming more rapid, but the majority still need to be more precise. Several studies were carried out to address these issues and improve metaheuristics-based approaches by combining the Apriori algorithm with several metaheuristics algorithms such as Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). The result of this combination gave birth to two approaches: GA-Apriori and PSO-Apriori. Consequently, after performing several studies on different database instances, the results revealed that the two approaches outperformed the Apriori algorithm in terms of runtime. PSO-Apriori also beats GA-Apriori in terms of both runtime and solution efficiency.

Technological breakthroughs have ushered in a revolution in a variety of industries, including agriculture, during the last several decades. This has given rise to what is now known as Agriculture 4.0, which emphasizes strategy and systems rather than the traditional obligations of the past. As a result, many human procedures have been replaced by a new generation of intelligent devices. Crop production management in Agriculture 4.0, on the other hand, poses a considerable challenge, particularly when it comes to prompt and accurate crop disease identification. Plant diseases are of special significance since they significantly reduce agricultural yield in terms of both quality and quantity. Deep learning neural network models are being utilized for early diagnosis of plant diseases in order to overcome this difficulty. These models can automatically extract features, generate high-dimensional features from low-dimensional ones, and achieve better learning results. In this research, we offer a joint solution involving image processing, phytopathology, and embedded platforms that intends to minimize the time necessary for human labor by leveraging AI to facilitate plant disease detection. We propose a learning-based PlantNet architecture for detecting plant diseases from leaf images, in which achieved about 97

Remote sensing target detection has been widely used in industries. In various application scenarios, complicated contexts may inhibit target identification and reduce detection accuracy, especially in multi-target detection tasks. In this paper, a new remote sensing target detection method based on structural reasoning is proposed to improve target detection performance by integrating inter-object relationships and scene information. Based on inter-object information, a relation structure graph is designed to reduce errors and missed targets. To establish contextual constraints, semantic is used as a prior information for Bayesian criterion based on scene information. Experiments conducted on HRRSD dataset show that the average accuracy of the proposed method is 10.7

Recommendation of services and applications based on user-data analytics is the most common approach to understanding user requirements. In this article, a novel technique for user recommendation is proposed and validated. The technique uses a Twitter Application Programming Interface (API) handle-based dataset for evaluating and computing the recommendations. The technique uses an open platform Graphical User Interface (GUI) for keyword categorization and building a reliable support system for query analysis. API driven queries from Twitter are cross-validated with labeling techniques and trending hashtags. Typically, the defined tweets are validated to build a Healthcare Record Indexing (HRI) data structure. The HRI is used to support the decision-making and recommendation of services of various healthcare applications and tweets. The technique has trained 750 datasets of categorized clusters with 150 000 tweets (dynamic) datasets from Twitter API. The technique has performed 92.68

With the development of recommendation technology, it is of great significance to analyze users' digital footprints on social networking sites, extract user behavior rules, and make a relatively accurate assessment of each user's needs, to provide personalized services for users. It has been found that the users' behavior on social networking sites has a great correlation with the user's personalities. The OCEAN model's five characteristics can cover all aspects of user personality. There are some shortcomings in the current micro-directional propagation method. This paper has improved the traditional collaborative filtering method and proposed a collaborative filtering method based on user clustering. The model first clusters the users according to their OCEAN model, and then it filters the users collaboratively in the cluster to which the user belongs with the collaborative filtering method based on an optimized singular value decomposition (SVD) recommendation algorithm, called the BiasSVD recommendation algorithm – to reduce the dimensionality of the data. Then it generates recommendations. Experiments show that clustering users' OCEAN models can improve the accuracy of recommendations. Compared with the entire user space, it greatly reduces the recommendation time, and effectively solves the cold start problem in micro directional propagation.

In this paper, an electromagnetism-like approach (EM) for solving the two-level hierarchical covering location problem (TLHCLP) is proposed. An EM metaheuristic is a powerful algorithm for global optimization that converges rapidly to the optimum. Therefore, it has the potential to solve this type of problem since movement based on the attraction-repulsion mechanisms, combined with the proposed scaling technique, directs EM to promising search regions. The fast implementation of the objective function and local search procedure for TLHCLP additionally improves the efficiency of the overall EM system. The proposed EM approach reaches all optimal solutions in a relatively short amount of computational time. EM also obtains high-quality solutions for large-scale problem instances that are out of reach for exact methods.

How to eliminate useless information in the vast network information and retain effective information is a problem that needs to be continuously explored in the field of deep learning. This paper conducts text classification on the network evaluation frequently encountered in daily life, mainly to screen out the meaningless comments published by Internet users, to have access to more useful information. In this paper, a text filtering model was constructed based on word vector and Long Short-Term Memory (LSTM) and improved by adding Deep Averaging Networks (DAN) and convolutional neural network (CNN). The major improvement of the LSTM & DAN model was to retain the original word vector information and to improve the accuracy of the text classification model without increasing hyperparameter and model structure complexity. The LSTM & CNN model mainly combines the advantages of convolutional neural network in exploring the deep information of text, which was an improvement over the original LSTM. It was proved by experiments that this improvement is meaningful. Compared with the shallow neural network, the accuracy has been greatly improved.