Theoretical considerations

Among big data analysis methods, the text mining technique derives formalized information from a large amount of non-formalized text data and constructs it in the form of new information, such as visualization and digitization. It can effectively extract and utilize meaningful information desired by the research when applied to text-based data, such as research literature. Therefore, this study attempted to quantitatively identify research trends using text mining and co-word analysis as research literature analysis methods. Visualization of Similarities (VOSviewer; Centre for Science and Technology Studies, Leiden University, Leiden, The Netherlands) is an analysis tool frequently used by many researchers for the analysis of bibliometric data, such as recent collaborative relationships between publications, journals, or researchers, and citation relationships between scientific terms (Heersmink et al., 2011; van Eck & Waltman, 2010; Waltman et al., 2010).

VOSviewer can create a two-dimensional (2D) map from network data, such as co-word and proximity matrices, and perform data clustering using its own clustering algorithm (Choi et al., 2011). It also includes an algorithm that identifies only the noun phrases from a text document after classifying and tagging the phrases in each sentence into verbs, adjectives, and nouns, thereby enabling analysis using the relevance score specified by the researcher. These functions extract words related to a specific research topic from the collected paper data and classify them into similar topics, and then analyze research trends by identifying the appearance frequency and association strength of each word.

The analysis principle of VOSviewer is to create an intuitive network map using nodes, appearance frequency, and association strength based on the number of edges, which can be explained in a conceptual analysis diagram (Fig. 1). For example, when A1 to A4 are assumed as authors or words and D1 to D3 as documents, A1 and A3 have a connection strength of 2 because both of them were used in two documents (D1 and D2).

This relationship analysis is mainly conducted in three steps. In the first step, a co-occurrence matrix is created for words that constitute the literature, and similarities between categories are calculated based on the matrix. The similarities are calculated using the association strength method in equation 1, one of the similarity calculation methods.

where C_ij is the number of simultaneous occurrences between items (words or papers) i and j. K_j and K_i are the total number of occurrences of i and j, respectively, and m is the population parameter. In the second step, a 2D map is created based on the similarities calculated in the first step. Items with high similarities are close, but those with low similarities are distant. In the third step, each parameter is clustered. Depending on the occurrence frequency, the graph is enlarged if the density of the parameters is high and reduced if it is low. In this study, keywords for each main research area were extracted to examine their association strengths and changes from the past to the present were investigated for the analysis of UAV-related research trends.

Preprocessing results

To identify UAV-related research topics, the co-occurrence frequencies of the words and phrases used in abstracts were analyzed. There were 13,359 occurrences extracted for domestic papers and 73,628 for international papers. Among them, words with similar meanings (e.g., unmanned, unmanned aircraft, aviation, and UAV) were excluded. Considering the size of the collected literature, the words that appeared in at least 10 domestic papers or 20 international papers were extracted in the first step. In the second step, a network map was created by setting Resolution to 1.5 and Min Cluster size to 5 in the clustering option for the visualization of the words that corresponded to the top 60% (VOSviewer setting), which acquired the relevance score as a keyword. A time-series network map to identify the time-series flow for research areas was created using the time information for the occurrence of the main keywords. The data before 2013 were excluded from the map because they did not meet the minimum criterion of the co-occurrence frequency.

UAV-related research trends

A total of 3,858 international papers from 1997 to 2017 were collected, and 745 domestic papers from 2002 to 2017 were collected (Table 1). In international countries, the number of papers increased by 7 per year from 1997 to 2008 and sharply increased by 89.8 per year from 2009 to 2017. In South Korea, the number of papers increased by 4 per year until 2013 and by 19.9 per year from 2014 to 2017 (Fig. 3). This appears to be because of the rapid growth of the private drone industry market, and the application range of drones expanded due to the improvement of hardware performance, miniaturization of related imaging devices, and the development of sensor technologies for drones. Considering the time at which the number of papers began to rapidly increase, there was a time difference of approximately five years between South Korea and international countries.

Research area classification results

The analysis results revealed that there were 173 main keywords for domestic papers and 477 for international papers. When clustering was performed based on this, domestic UAV-related studies could be classified into seven key areas compared to eight areas for international studies. In addition, a time-series network map was created based on the keyword appearance time to identify changes in the time series for each research area.

As for international studies, 99 co-words were derived in Cluster1, a research area related to “remote sensing/data analysis”. The cluster represented the largest proportion in the entire research area. The word that appeared most frequently was “monitoring”, and the cluster was highly associated with “tree”, “plant”, “vegetation”, and “species” (Fig. 4).

Research on UAV remote sensing is similar to research that uses satellites and aerial imaging in terms of methodologies, but this study was conducted with a focus on the characteristics of UAV images that can acquire more precise spatial information than the existing data. The publication of related research literature has shown an increase since 2014, and research has been conducted most actively along with the Cluster4 area (Fig. 5).

In Cluster2, a research area related to “mission performance”, 77 co-words were derived and the word that most frequently appeared was “simulation”. This cluster was highly associated with “path”, “constraint”, “obstacle”, and “collision”. As UAV is a research area on flight paths, control, obstacles, and collisions for performing UAV missions, it was dominated by experimental studies using software and hardware. Considering the publication time and number of papers, research was most active before 2014 and has been decreasing recently. The stabilization of the corresponding technologies appears to be the main cause (Table 2).

Cluster3 is a research area related to “flight control”. A total of 76 co-words were derived and the word that most frequently appeared was “control”. This cluster was highly associated with “trajectory”, “dynamic”, “stability”, and “scheme”. Studies on trajectory, dynamics, stability, and design, which are related to the UAV control performance, corresponded to this cluster. This cluster is in the center of Cluster2 and Cluster5, which are highly associated research areas where research has been continuously conducted.

In Cluster4, a research area related to “data collection/processing”, 76 co-words were derived. This cluster is close to Cluster1, indicating a high association in terms of contents. While Cluster1 represents applied research for various areas, Cluster4 is a research area for improving the basic functionality of UAV images. The word that most frequently appeared was “image”, which was used with “quality”, “accuracy”, “site”, “map”, and “point cloud”. This cluster was dominated by studies on software processing, such as UAV image quality, accuracy, maps, destinations, mapping, and point cloud data. Active research has been conducted in this area.

Cluster5 is a research area related to “airframe performance”. Sixty-nine co-words derived and the word that most frequently appeared was “design”. This cluster was highly associated with “aircraft”, “angle”, “degree”, “wing”, and “landing”. As UAV is a research area related to the flight performance according to the airframe structure, research was most active around 2010 but has been decreasing.

Cluster6 is a research area related to the “UAV technology/industry trend”. Thirty-five co-words derived and the word that most frequently appeared was “tool”. This cluster was highly associated with “year”, “precision”, “variety”, and “opportunity”. It was dominated by studies with a focus on the precision and various applications of the data acquired by UAVs.

Cluster7 is a research area related to “navigation/location-based technology”. Thirty-one co-words were derived and the word that most frequently appeared was “position”. This cluster was highly associated with “velocity”, “navigation”, “GPS”, and “localization”. Research on the application, accuracy verification, and improvement of location-based technologies, such as velocity, navigation, and GPS, among the functions required for UAV operation, corresponds to this cluster.

Cluster8 is a research area related to “radio wave/radar”. Thirteen co-words were derived and the word that most frequently appeared was “pattern”. This cluster was highly associated with “band”, “antenna”, “validity”, and “processing”. It is a research area on the acquisition, processing, and analysis of radar images that can only be obtained by large UAVs.

As for domestic studies, 40 co-words were derived in Cluster1, a research area related to the “UAV technology/industry trend”. The cluster represented the largest proportion in the entire research area. The word that most frequently appeared was “aircraft”, which was used to indicate UAVs. The words that exhibited a high association were “Korea”, “industry”, “regulation”, “law”, and “trend”. In South Korea, research on regulations and laws for UAVs began in 2013, and studies on the UAV industry and technology trends have been increasing recently (Figs. 6, 7).

Cluster2 is a research area related to “data collection/processing”. Thirty-five co-words were derived and the word that most frequently appeared was “image”. The words that exhibited a high association were “accuracy”, “map”, “field”, “survey”, and “error”, which were similar to the co-words in Cluster4 of the international studies. This cluster was dominated by studies with a focus on the utilization of UAV images, such as location accuracy, height, and mapping. Such studies began to increase in 2015 and have been actively conducted (Table 3).

Cluster3 is a research area related to “flight control”. Twenty-eight co-words were derived and the word that most frequently appeared was “performance”. The words that exhibited a high association were “design”, “control”, “simulation”, and “stability”. Simulation and experimental studies related to the UAV control performance correspond to this cluster. Such studies were active around 2014 and related studies have decreased recently.

Cluster4 is a research area related to “airframe performance”. Twenty-three co-words were derived and the word that most frequently appeared was “test”. The words that showed a high association were “power”, “configuration”, “wing”, “hale”, and “shape”. This cluster was dominated by test studies on wings, batteries, and airframe geometry, which are related to endurance.

In Cluster5, a research area related to “remote sensing/data analysis”, 21 co-words were derived. The word that most frequently appeared was “monitoring” as in Cluster1 of the international studies where the corresponding studies were similar. The words that exhibited a high association included “estimation”, “coefficient”, “imagery”, “height”, “classification”, and “NDVI”. This cluster was dominated by papers on vegetation. Such papers have increased since 2015.

Cluster6 is a research area related to “reconnaissance technology”. Fifteen co-words were derived and the word that most frequently appeared was “algorithm”. The words that exhibited a high association were “mission”, “surveillance”, and “reconnaissance”. Studies on reconnaissance and surveillance using UAVs correspond to this cluster. In this research area, which is distinguished from the international analysis results, the contents of the research papers were mainly published for military purposes, but most of the research results focused on technology trends and utilization. Therefore, this cluster is determined to be similar to Cluster1.

Cluster7 is a research area related to “navigation/location-based technology”. Eleven co-words were derived and the word that most frequently appeared was “signal”. The words that showed a high association were “navigation”, “flight path”, and “ground control system”. Studies related to location-based technology and the accuracy of the UAV route correspond to this cluster. This cluster exhibited a low association with other research areas and was observed in limited time periods.

When the domestic and international research trend analysis results were summarized, it was found that they were generally clustered into areas with similar meanings, but the year the keywords appeared differed, which indicated research topics, association strength between words, and diversity. For example, basic studies on UAV technology trends and utilization were abundant in South Korea, but studies on data collection and analysis through UAV remote sensing technology were most abundant in international countries.

Fig. 8 compares changes in main domestic and international research areas for 17 y between the two periods. The main international research areas between 2000 and 2013 were found to be “mission performance”, “airframe performance”, and “location-based technology”. After 2014, research areas related to “remote sensing/data analysis”, “data collection/processing”, and “UAV technology/industry trend” sharply increased. This change appears to be the result of the stabilization of UAV-related basic technologies and the gradual increase in demand for application areas. On the other hand, the main domestic research area between 2000 and 2013 was found to be “airframe performance”, and other studies were not significant. Since 2014, research has been actively conducted in all areas and a tendency to follow the international research areas was observed. Research is expected to be expanded in various areas in the future even though the research area related to “remote sensing/data analysis” shows a time difference of approximately three to four years.

Ecological environment research area derivation using UAVs

The research area related to the ecological environment is included in “remote sensing/data analysis”, which corresponds to Cluster1 for international studies and Cluster5 for domestic studies. The words that appeared were “monitoring”, “plant”, “crop”, “classification”, “forest”, “temperature”, “tree”, “species”, “Normalized Difference Vegetation Index (NDVI)”, and “Leaf Area Index (LAI)”, which were mostly associated with plant ecology (Table 4). These words became the keywords of research topics that attracted attention recently because the advancement in UAV optical sensor technology and the development of image processing technology made it possible to acquire data that could not be obtained from the existing spatial information. The fact that related domestic papers have recently increased in a similar manner to international research trends indicates the recognition of the importance for future research in this area.

In conclusion, UAVs have attracted attention as the key industry of the fourth industrial revolution era because they integrate innovative technologies, such as aviation, information communication, sensors, and software, and can be combined with various areas. Despite this potential, research that utilizes the data acquired through UAV systems is in the preliminary stage. In particular, studies on spatial information acquisition and analysis methods in the ecological environment area are extremely insufficient. This study examined domestic and international research trends using the text mining analysis method and found that active attempts have been made to apply various research methodologies to UAVs. For example, the shape of the canopy gap was identified through ultra-high-resolution UAV images to analyze species diversity for forest vegetation, and such images were used to collect time-series vegetation index data for research on phenology. Moreover, the ground surface temperature images of vegetation distribution areas were acquired for their wide utilization in ecosystem research, such as habitat environment analysis.

The keywords that appeared in such domestic and international research literature were “monitoring”, “classification”, “height”, “temperature”, and “NDVI”. Among them, “monitoring” exhibited the highest appearance frequency. This is because it was used as a gerund along with subwords. For example, it was used as keywords that represent research topics, such as “plant monitoring” and “tree monitoring”. “Classification” is a word that appeared in studies related to the extraction of ground surface information, such as land cover and vegetation classification, using matching images. “Height” was found in studies that utilized tree height information to identify the growth status in relation to farmlands or crop cultivation. “Temperature” is a keyword in studies that analyzed the habitat environments of animals and plants or identified individuals through precise ground surface temperature image data. NDVI is the vegetation information utilized in various research types, such as forests and farmlands, and it most frequently appeared in studies that used near-infrared cameras.

While research has been conducted on various topics internationally, UAV-related domestic research is in the early stage. Although preliminary studies were conducted at different target sites on different topics, studies that were examined in terms of accuracy verification and utilization based on the field data are insufficient.

Therefore, to increase UAV utilization in the ecological environment area, it is necessary to increase the accuracy of testing and calibration methods considering various factors that affect homogeneous data acquisition, such as weather, airframe performance, and image processing methods, in processes from data acquisition to the derivation of results. Follow-up research on this topic will increase UAV utilization in the ecological environment area.