《1 Engineering research hotspots and engineering research focus》

1 Engineering research hotspots and engineering research focus

《1.1 Development trends of engineering research hotspots》

1.1 Development trends of engineering research hotspots

The top 10 engineering research hotspots in the information and electronic engineering field are summarized in Table 1.1.1. They cover the subfields of electronic science and technology, optical engineering and technology, instrument science and technology, information and communication engineering, computer science and technology, control science and technology, and other disciplines. Among them, “Multi-object tracking technologies,” “Characteristics and analyses of neural networks/neural network modeling and fuzzy logic,” “Robust adaptive boundary control,” “Ultra-wideband slot antennas,” “Dynamic network visualization,” “Global optimization algorithms,” “Multi-criteria decision making,” “Reversible data-hiding schemes,” “High-utility itemset mining algorithms,” and “Optical coherence elastography” are in-depth research topics of traditional research.

The numbers of core papers in each engineering research hotspot published annually from 2011 to 2016 are shown in Table 1.1.2.

(1) Multi-object tracking technologies

In recent years, multi-object tracking has emerged as an important science and technology. It is related to a variety of disciplines, including stochastic statistics, mathematical optimization, image processing, pattern recognition, artificial intelligence, and automatic control. Multi-object tracking has been widely used in military applications, especially in the application of airborne detection and remote warning. The current issues in multi-object tracking include occlusions, the randomness of the number of objects, complex backgrounds, and real-time requirements. The focus of overseas research has shifted from multi-object tracking technology using a single sensor to multi-object tracking with multiple sensors. With the rapid movement of objects and the changing environment of multi-object tracking technology, tracking the maneuvering object has attracted a renewed interest. Another research hotspot is the implementation of artificial neural networks to achieve multi-object tracking. Owing to the potential selforganization and self-learning abilities of artificial neural networks, this will be a promising research direction for multi-object tracking.

(2) Global optimization algorithms

According to the optimization objective, the global

《Table 1.1.1》

Table 1.1.1 Top 10 engineering research hotspots in information and electronic engineering

《Table 1.1.2》

Table 1.1.2 Annual number of core papers belonging to each of the top 10 engineering research hotspots in information and electronic engineering

optimization algorithms are divided into singleobjective, multi-objective, and hierarchical multi-objective optimization problems. According to the data type, they are divided into discrete, continuous, and hybrid optimization problems. Because global optimization problems often have many local optimal solutions, they are difficult to solve with the classical optimization method. Particularly in the big data environment, there may be a huge number of local optimal solutions, which creates a considerable challenge in the search for a global optimal solution. In particular, no effective global judgment criterion has been determined. Consequently, the advantages and disadvantages of different local optimal solutions are difficult to evaluate, which hinders the unification of the global optimization objective. Global optimization problems are frequently encountered in data mining, image processing, machine learning, artificial intelligence, and other fields. Different global optimization algorithms are used for different data characteristics and targets.

(3) Robust adaptive boundary control

Robust adaptive control concerns systems with uncertainties and has broad application areas, such as robotics and flight control. Robust adaptive boundary control is a rather large category and includes boundary disturbance attenuation, nonlinear active disturbance control, adaptive boundary control, and sliding mode control.

The theory of boundary measurement and control was developed within the robust adaptive boundary control research domain. It is critical to target the most challenging engineering problems that arise in real practical applications and to form novel solutions using boundary measurement and control methodologies. The challenging areas in which robust adaptive boundary control is needed include—but are not limited to—active control of combustion instabilities, structural vibration control of aerospace vehicles, active flow control, pinning control of large-scale networks, boundary measurements and related inverse problems in biologic tissues, and boundary manipulation physics and control system technologies.

(4) Multi-criteria decision making

Multi-criteria decision making can be divided into two classes, multi-attribute decision making and multiobjective decision making, based on whether the decision scheme is finite or infinite. A current means of multicriteria decision making is to merge the effectiveness of all attributes to obtain the full effectiveness and then to rank the schemes. In 1965, Zadeh proposed the fuzzy set theory. In 1970, Bellman and Zadeh introduced the fuzzy set theory in multi-criteria decision making and proposed the fuzzy multi-criteria decision making (FMCDM) model to resolve the uncertainty problem in actual decision making; this model has become a popular research direction in the current multi-criteria decision making research hotspot.

(5) Reversible data-hiding schemes

Reversible data hiding is an information-hiding technology that can fully restore the source data after hidden data is correctly extracted. Reversible data hiding is used mainly in covert communications, copyright protection, and content integrity authentication. Reversible data hiding is classified into three categories: ① single-pixel hiding and pixel group hiding, ② spatial domain hiding and transform domain hiding, and ③ even hiding and uneven hiding. At present, research on pixel group hiding, uneven hiding, and difference expansion is the most popular.

Anti-attack is another research topic that must be investigated simultaneously with data hiding. Attackers aim to destroy original source data or extract hidden information; therefore, anti-attack capabilities must rely on other information encryption methods.

(6) High-utility itemset mining algorithms

The high-utility itemset mining algorithms have played a more intuitive and direct-supporting role in solving the decision-making problem in some practical application scenarios. In 2004, Yao et al. proposed the concept of highutility itemset mining. Other researchers have proposed a number of high-utility itemset mining algorithms, including the incremental high-utility pattern (IHUP) algorithm, utility pattern growth (UP-growth) algorithm, and high-utility itemset miner (HUI-miner) algorithm.

Although high-utility itemset mining has made remarkable achievements, challenges remain in the following research areas: data mining of columnar storage, development of appropriate standards for measuring high-utility itemsets, visual mining, parallel mining of high-utility itemsets, and mining of high-utility itemsets in network environments.

(7) Optical coherence elastography

Optical coherence elastography (OCE) is of great importance in the research of microstructure imaging of biological tissue. According to the mode of excitation, OCE can be classified as sound surface-wave propagation OCE, shear wave propagation OCE, acoustic radiation force OCE, pressure-type OCE, and magnetic induction OCE. OCE offers a special advantage in measuring the unique organizational level of biomechanics at the micron level, cell level, and even the molecular level. The method is a new nondestructive approach to evaluating tissue elasticity and is superior to ultrasonic imaging and magnetic resonance imaging. Its development process is the same as those of other means of elastic imaging, from static imaging to dynamic imaging, and from low resolution to high resolution. In the future, the image resolution, imaging depth, and imaging velocity of the method are expected to be gradually improved.

(8) Ultra-wideband slot antennas

The ultra-wideband slot antenna has a basic antenna form in which radiation is produced through grooving on metal and excitation with a coaxial cable, microstrip cable, or waveguide. The slot antenna can replace the oscillator antenna in the microwave section, solving problems that the oscillator antenna is too small to make and feed. The wideband characteristics of the circular and elliptic slot antennas are well known. A slot antenna can achieve ultra-wideband bandwidth and its bandwidth depends on the shape of the slot. For example, the length-to-width ratio of the slot has a significant impact on the bandwidth of an antenna with a wide rectangular slot, and the longto-short axis ratio has great influence on the bandwidth of the elliptical slot. Improvement of the feed network can greatly improve the working bandwidth.

(9) Dynamic network visualization

In a dynamic network structure, the nodes and edges of a network evolve over time. A dynamic network can be observed and analyzed by visualization, and important insights can be obtained. In an existing network, some features can be automatically analyzed with traditional data analysis methods. However, the greatest advantage of combining traditional methods and visualization is that useful patterns can be determined by using large amounts of data and further observation with traditional methods. In dynamic network visualization, more attention must be paid to the heterogeneity of the key topology and the time dimension of the network data. Data visualization of large-scale networks is a very challenging topic and a well-known problem.

(10) Characteristics and analyses of neural networks/ neural network modeling and fuzzy logic

A neural network is a mathematical model that imitates the behavior of animal neural networks and processes information in a distributed and parallel way. The combination of modeling based on neural networks and fuzzy logic can improve the generalization ability and efficiency of the model. Currently, the main technological development orientations are: first, improving the accuracy of fuzzy systems by adjusting fuzzy logic membership functions based on neural networks; second, using fuzzy logic to improve the learning ability of the neural network and appropriate heuristic control logic to control the learning process to increase the training efficiency and reduce the demand for data quality. Based on these methods, a combination of the fuzzy system and the neural network is constructed to improve the generation quality and the ability of the model to acquire special knowledge. A neural network is used to approximate a fuzzy logic system, which reduces the difficulty of computer implementation and improves engineering efficiency.

With the development of deep learning technology, the theory and technology of neural networks have advanced rapidly. Additionally, the integration with other disciplines has become more abundant in recent years, which has further promoted the development of neural network technology.

《1.2 Understanding of engineering research focus》

1.2 Understanding of engineering research focus

1.2.1 Multi-object tracking technologies

Multi-object tracking technology has been widely used in video surveillance, autonomous driving, humancomputer interaction, virtual reality, and augmented reality, and has significant practical value in military and civilian fields. Multi-object tracking has been greatly enhanced by the rapid development of machine learning methods, the continuous improvement of hardware performance, and increasingly rich data sets. To meet the demand for a high-performance and robust multi-object tracking algorithm for practical applications, a series of challenging scientific and technical problems have been introduced, including the appearance modeling of objects, visual representation and feature selection, occlusions, data association algorithms, probability inference between state changes in video sequences, and decision optimization. To address the above challenges, innovative research is needed based on the rapid development of modern artificial intelligence technology through the combination of theory and practice, model and algorithm, and new ideas and changes in model systems. At present, the main hotspots of this research frontier can be summarized as follows.

(1) Multi-object initialization. The development of a full-time multi-object tracker with speed and accuracy but without limitations on the type of objects being tracked remains a frontier topic.

(2) Appearance modeling for multiple objects. In visual recognition, effective feature representation remains an interesting and challenging issue. Multi-object tracking requires not only the design of robust visual representations, but also accurate statistical measurements to estimate the similarities between different observations.

(3) Motion modeling for multiple objects. The movement of an object in a scene is usually erratic and it is difficult to model acceleration or stopping. The effective multi-object tracking algorithm must design and implement a nonlinear motion model.

(4) Combination of online and offline tracking. The implementation of these two dissimilar methods to obtain a simpler and more effective multi-object tracking algorithm remains a challenging topic.

(5) Object association algorithm. Overlapping and blocking between multiple objects is an inevitable fatal problem in multi-object tracking. The establishment of an effective way to deal with overlapping and occlusion is an important future development direction.

There are 61 core papers focusing on “Multi-object tracking technologies” (Table 1.2.1). The top five countries are the USA, China, Switzerland, Germany, and Korea. Papers published by authors from the USA accounted for 56.25%, more than half, and two of the three consistently cited papers are published by the USA. China ranked the second with core papers accounting for 20.83% of the total number of papers. According to Table 1.2.2, the top five institutions in paper output are the University of Southern California (Univ So Calif), the Chinese University of Hong Kong (Chinese Univ Hong Kong), Darmstadt University of Technology in Germany (Tech Univ Darmstadt), the Swiss Federal Institute of Technology in Zurich (ETH), and the University of Central Florida (Univ Cent Florida), two of which are in the USA and one is in China. From the network analysis of collaboration among researchers in different countries or regions, shown in Figure 1.2.1, Figure 1.2.2, Table 1.2.3, and Table 1.2.4, the USA occupies the core position. The cooperation among published researchers in the top four countries or regions is also closer.

Based on the results of this analysis, it is clear that the USA, regardless of the number of papers or the collaboration with other countries or regions, is the leader in this research focus, while China is following closely to 

《Table 1.2.1》

Table 1.2.1 Major producing countries or regions of core papers on the engineering research focus “Multi-object tracking technologies”

《Table 1.2.2》

Table 1.2.2 Major producing institutions of core papers on the engineering research focus “Multi-object tracking technologies”

《Figure 1.2.1》

Figure 1.2.1 Collaboration network of the major producing countries or regions of core papers on the engineering research focus “Multi-object tracking technologies”1

1 In the figure, the nodes refer to the countries or regions, the size of the nodes refers to number of papers, the connecting line between nodes refers to papers published based on research cooperation, and the thickness of the connecting line indicates the number of papers based on research cooperation. These are the same in full text.

《Figure 1.2.2》

Figure 1.2.2 Collaboration network of the major producing institutions of core papers on the engineering research focus “Multi-object tracking technologies”

《Table 1.2.3》

Table 1.2.3 Major producing countries or regions of core papers that are cited by core papers on the engineering research focus “Multiobject tracking technologies”

《Table 1.2.4》

Table 1.2.4 Major producing institutions of core papers that are cited by core papers on the engineering research focus “Multi-object tracking technologies”

the USA, which cooperates closely with other countries or regions.

1.2.2 Dynamic network visualization

Because the data produced in most computerrelated and communication fields have an additional time dimension compared with traditional data, the good organization of the data and their presentation with efficient visualization methods are valuable for understanding and analyzing their inner structure. Solving this problem is of great significance to the communication industry and academic research. Scholars and experts in the field of visualization and visual analysis worldwide are trying to solve this problem.

Among the scholars studying dynamic network visualization, those in the USA produce 31.91% of the core papers in this field, ranking the first; they are followed by researchers in Germany, the Netherlands, France, China, and other countries or regions. The core papers produced in the top 10 countries or regions are cited more than 20 times (Table 1.2.5). Peking University, University of Toulouse, the French Civil Aviation University (ENAC), and University of Groningen are the top four institutions. Four papers are produced by each of these four institutions, with more than 90 citations, and a proportion of citations of more than 9% (Table 1.2.6). The collaboration networks between the major countries or regions and the major institutions on the engineering research focus “Dynamic network visualization” are shown in Figure 1.2.3 and Figure 1.2.4.

Regarding the core papers that are cited by core papers, Germany, Switzerland, and China are the top three countries based on the country or region statistics, with no less than three core papers that are cited by core papers, and the proportions of those papers are higher than 15% for each of the three countries in the list (Table 

《Table 1.2.5》

Table 1.2.5 Major producing countries or regions of core papers on the engineering research focus “Dynamic network visualization”

《Table 1.2.6》

Table 1.2.6 Major producing institutions of core papers on the engineering research focus “Dynamic network visualization”

《Figure 1.2.3》

Figure 1.2.3 Collaboration network of the major producing countries or regions of core papers on the engineering research focus “Dynamic network visualization”

《Figure 1.2.4》

Figure 1.2.4 Collaboration network of the major producing institutions of core papers on the engineering research focus “Dynamic network visualization”

1.2.7). Hong Kong University of Science and Technology ranks the first among institutions with three papers with a proportion higher than 12%. It is followed by the Max Planck Institute for Human Cognitive and Brain Sciences, Nanyang Technological University, and the Swiss Federal Institute of Technology, with two papers each and a proportion of 8% for each institution (Table 1.2.8).

Currently, research on “Dynamic network visualization” focuses on incremental rendering on the node-connection graph, especially on some types of graphs, such as trees, strings, parallel graphs, and directed acyclic graphs. Many scholars aim to use node topology to visualize a dynamic network in the 2D plane, and visual forms with a more visual metaphor effect, such as 3D graphs, have been proposed in the literature. In addition to rendering, interesting work is being performed on the scalability of visual dynamic networks. Little information exists in the literature on the methodology for visualizing large dynamic network data sets. With an increasing amount of data, the nodes overlap and the traditional method inevitably produces complexity in visual computing. Animation is a commonly used approach to explaining dynamic network changes, and the tradeoff between data processing speed and rendering performance should be considered. Efforts devoted to these research topics will be significant for understanding the structure and relationship of dynamic network data.

1.2.3 Characteristics and analyses of neural networks/ neural network modeling and fuzzy logic

Recently, with the development of the Internet and big data, the global demand for artificial intelligence 

《Table 1.2.7》

Table 1.2.7 Major producing countries or regions of core papers that are cited by core papers on the engineering research focus “Dynamic network visualization”

《Table 1.2.8》

Table 1.2.8 Major producing institutions of core papers that are cited by core papers on the engineering research focus “Dynamic network visualization”

technology has become enormous. Neural networks, particularly deep neural networks that represent depth learning, have promoted the development and application of artificial intelligence in technology. The development of neural networks has been an important issue in practical applications, such as Google’s AlphaGo, which further demonstrates the outstanding performance of depth learning in artificial intelligence. From the data presented in Table 1.2.9 and Table 1.2.10, we can see the technical demands and development potential of different countries or regions in this field. The combination of fuzzy logic and neural networks utilizes and balances the advantages and disadvantages of these approaches and is another important research field.

From keywords of the core papers such as neural network, deep learning, and fuzzy control, the leading edges of the research field include: optimizing neural networks, mainly by optimizing the learning algorithm; optimizing self-learning structures, mainly the model’s self-adjustment ability; and optimizing fuzzy rules.

Among the leading issues of the combination of neural network learning algorithms with fuzzy logic are structure learning and parameter learning. Rule learning includes the heuristic search method, the fuzzy grid method, the tree partition method, and the learning algorithm based on fuzzy clustering. The research in this field can further

《Table 1.2.9》

Table 1.2.9 Major producing countries or regions of core papers on the engineering research focus “Characteristics and analyses of neural networks/neural network modeling and fuzzy logic”

《Table 1.2.10》

Table 1.2.10 Major producing institutions of core papers on the engineering research focus “Characteristics and analyses of neural networks/neural network modeling and fuzzy logic”

improve the performance and training speed of neural networks, and the combination of distributed systems can enhance the generation speed of model parameters. The leading countries and regions in this field of research include the Chinese mainland, Taiwan of China, the USA, and Canada.

The optimization of structure self-learning includes mainly connection design methods and evolutionary design methods. Connection design methods include the supplementary algorithm and cut algorithm. The research in this field can reduce redundant parameters and the volume of the neural network and provides a feasible method for optimized neural network modeling, which has advanced neural network engineering. The leading countries and regions in this research field include the Chinese mainland, the USA, Canada, and Australia.

Fuzzy logic rules are often obtained empirically and have poor precision. Fuzzy logic rules with the self-adaptive characteristics of neural networks solve the network black box problem. Although scholars have proposed some methods to further promote the development of this field, no breakthrough has been achieved. In addition, other research areas, including the development of a probabilistic graph model and convex optimization theory, have made some progress in explaining the neural network black box problem by theoretically determining the characteristics of neural networks to some degree. At present, the leading countries and regions in this field include the Chinese mainland, the USA, Canada, and Taiwan of China.

Besides combining fuzzy logic and neural networks, other research has been conducted on overcoming the limitations of neural networks, such as combining the generative adversarial net of game theory to further reduce the demand quality of network data. Additionally, with the development of deep neural networks, model parameters and the compression problem have become important research topics, and the study of the neural network itself is still in progress. In recent years, the convolution neural network, recurrent neural network, and long short-term memory network have become the focus of neural network research and been implemented in engineering. Meanwhile, some specific structures, like the activation function, are being constantly developed. With the continuous progress of artificial intelligence technology, the field will be further integrated with other research areas and is expected to have wide implications and produce new technologies.

According to the network map that illustrates the collaboration among the key engineering research associations, the currently major research countries and regions are the China, Taiwan of China, Saudi Arabia, and India (Figure 1.2.5 and Figure 1.2.6). China ranks the first in the number of core papers and the number of core papers that are cited by core papers (Table 1.2.11), which demonstrates that China has become the key research country in this field. However, the number of institutions producing core papers that are cited by core papers ranks

《Figure 1.2.5 》

Figure 1.2.5 Collaboration network of the major producing countries or regions of core papers on the engineering research focus “Characteristics and analyses of neural networks/neural network modeling and fuzzy logic”

《Figure 1.2.6》

Figure 1.2.6 Collaboration network of the major producing institutions of core papers on the engineering research focus “Characteristics and analyses of neural networks/neural network modeling and fuzzy logic

lower (Table 1.2.12), which shows that China must further optimize the allocation and improve the concentration of research resources in this field. In addition, China and some countries or regions that are active in this technical field rarely cooperate. Therefore, promoting academic exchange and cooperation should also be a focus of the development efforts in this technical field in China.

《2 Engineering development hotspots and engineering development focus》

2 Engineering development hotspots and engineering development focus

《2.1 Development trends of engineering development hotspots》

2.1 Development trends of engineering development hotspots

The top 10 engineering research hotspots, summarized by the Field Group of Information and Electronic Engineering, are shown in Table 2.1.1. They cover the subfields of electronic science and technology, optical engineering and technology, instrument science and technology, information and communication engineering, computer science and technology, control science and technology, and other disciplines. Among them, “Wireless mobile communication technology and equipment,” “Vehicle autonomous-driving technology and its application,” “Laser light source technology,” “Image recognition and visual interaction technology,” “Cloud computing and virtualization,” and “Unmanned aerial vehicle” are areas of further study in traditional research. The annual numbers of core papers published in each hotspot from 2011 to 2016 are shown in Table 2.1.2.

(1) Image recognition and visual interaction technology

Image recognition and visual interaction technology

《Table 1.2.11》

Table 1.2.11 Major producing countries or regions of core papers that are cited by core papers on the engineering research focus “Characteristics and analyses of neural networks/neural network modeling and fuzzy logic”

《Table 1.2.12》

Table 1.2.12 Major producing institutions of core papers that are cited by core papers on the engineering research focus “Characteristics and analyses of neural networks/neural network modeling and fuzzy logic”

《Table 2.1.1》

Table 2.1.1 Top 10 engineering development hotspots in information and electronic engineering

《Table 2.1.2》

Table 2.1.2 Annual number of core patents belonging to each of the top 10 engineering development hotspots in information and electronic engineering