Ontology Construction and Evaluation of UAV FCMS Software Requirement Elicitation Considering Geographic Environment Factors

The quality of unmanned aerial vehicle flight control and management system (UAV FCMS) software is crucial to guaranteeing the quality of UAVs. Software requirement elicitation (SRE) is an important part of the UAV FCMS software development process. However, this activity suffers from ambiguity, heterogeneity and incompleteness; furthermore, because the use of UAVs is closely related to their geographic environment, geographic environment factors must be fully considered when conducting UAV FCMS SRE activity. In the knowledge engineering community, an ontology is an explicit specification of a conceptualization. Introducing the ontology method into the SRE process is an effective way to solve the above problems. This paper creates a UAV FCMS SRE ontology (SREO) based on domain knowledge and empirical data, as well as a geo-ontology based on geographic information metadata. Then, the paper integrates the above two ontologies into a new ontology. The goal of ontology integration is to analyze ontology concepts by adopting a hybrid ontology mapping method. The specific process analyzes the semantic similarities between the concepts of two ontologies and then decides whether to use a description logic (DL) strategy based on the analysis results. When the corresponding conditions are satisfied, the DL strategy is used to perform both direct and transitive reasoning for the relationships to achieve the ontology mapping, and the ontology integration is eventually implemented. Finally, this paper uses a criteria-based ontology evaluation approach to evaluate the quality of the newly integrated ontology. The evaluation results show that the UAV FCMS SREO considering geographic environment factors exhibits high quality. Further engineering practices also show that the SRE activities and the generated software requirement specifications (SRSs) exhibit a large increase in quality. Through the above activities, improvements to both the quality and reliability of UAV FCMS software can be achieved.

waypoint may result in a UAV flight path error'', ''when a UAV is in a state of dead reckoning navigation, there may be an unsafe situation when flying at a fixed altitude in plateau'', etc. Clearly, these problems are directly related to UAV flight control and management systems (UAV FCMSs) since the quantities relevant to the geographic environment are often used as inputs to the UAV FCMSs. In addition, the special functions and tasks of UAVs give them remarkable domain characteristics. Moreover, the openness and nondeterminism of UAV operating environments, the complexity of the interactions between system components and environments, and the unpredictability of operating conditions and scenarios all intensify the knowledge-intensive development trend of UAVs and make their effectiveness highly dependent on the quality of domain knowledge.
UAV FCMSs play a decisive role in the flight performance, reliability and safety of UAVs. Moreover, software is an integral part of UAV FCMSs. Therefore, ensuring and improving the quality of UAV FCMS software is crucial to guaranteeing the quality of UAVs. Software requirement elicitation (SRE) is the most critical knowledge-intensive activity in a software development process; however, implementing effective requirement elicitation and obtaining correct, complete, consistent, and unambiguous requirement specifications remains a problem that plagues system analysts and software developers. These issues also exist in UAV FCMS SRE activities and have a significant impact on their quality. An important reason for the above problems is the lack of an effective knowledge sharing bridge between system developers and domain users [3]. In addition, the increasing scale and complexity of software systems add to the difficulties in comprehension and development. Moreover, different teams with multiple-views and multi-paradigm development methods are widely used in the development of such complex software systems, which increases the heterogeneity of software requirement specifications (SRSs) and results in inconsistent and ambiguous SRSs [3]. A knowledge-based requirement elicitation method can be used to solve the above problems; its purpose is to use domain analysis and experience to help software system stakeholders understand the application domain and define requirements. The key is to model the domain knowledge as a shareable knowledge framework. Under this framework, domain users can more easily and conveniently express their needs, while the domain developers can understand the requirements more accurately. Moreover, the heterogeneity brought about by multiple viewpoints and paradigms can be minimized. In the knowledge engineering community, an ontology is a formal and explicit specification of a shared conceptualization [4], [5]. Therefore, introducing the ontology method into the SRE process is an effective way to solve the above problems. By adopting the ontology method, the requirement knowledge can be expressed as an ontological concept and association; therefore, it is clear, complete, and consistent and is conducive to the sharing and reuse of knowledge. Literature [6] designed an ontology in a case study for co-simulation in a model-based system engineering tool-chain. They argued that an ontology refers to content about the types of objects, their properties, and their relationships, which represent domain-specific knowledge. Literature [2] studied a software requirement error pattern (SREP) based on an ontology and illustrated the application process with a certain type of UAV FCMS software, but they did not explicitly consider the influence of geographic environment factors on software function realization and reliability assurance. Because UAVs access knowledge in multiple domains, it is necessary to integrate the ontologies of different domains to achieve a relative completeness of knowledge.
This paper proposes a hybrid ontology mapping method. Based on traditional similarity calculations, this paper adopts the deductive reasoning of a description logic (DL) strategy to perform both direct and transitive reasoning for the relationships to achieve the ontology mapping, and ultimately implement ontology integration. By integrating part of the geographic information of a geo-ontology into a UAV FCMS SRE ontology (UAV FCMS SREO), the semantics of the ontology are enriched; thus, the completeness of ontology knowledge is enhanced. Furthermore, the ontology mapping can also eliminate the ambiguity and heterogeneity to some extent. On this basis, this paper adopts a criteria-based evaluation approach to evaluate the quality of the integrated ontology, including ontology validation and ontology verification. Ontology validation checks if the correct ontology has been built, whereas ontology verification checks if the ontology has been built correctly [7]. The rest of this paper is divided into the following sections: section 2 presents the state of the art in ontology integration and evaluation. Section 3 describes the construction of the UAV FCMS SREO based on domain knowledge [8]- [10], industry standards [11] and experience (software requirement errors); moreover, it describes the elicitation of the related concepts of the UAV FCMS SRE based on the literature [12], [13] and the construction of the geoontology. Section 4 describes the integration of the UAV FCMS SREO and the geo-ontology and presents the hybrid ontology mapping method combining the semantic similarity calculation [14] with DL [15], [16]. Section 5 presents a case study involving the implementation of the hybrid ontology mapping method and a quality evaluation of the newly integrated ontology in terms of ontology validation and ontology verification. Ontology validation is achieved by applying two validation methods [7]. The first is the ontology content evaluation, and the second is answering competency questions. Ontology verification is achieved using two methods, also [7]. The first is the ontology taxonomy evaluation, and the second is the improved FOCA methodology [17]. This section also shows the results of engineering applications. Finally, section 6 concludes the study.

II. RELATED WORK
The current state of the art of ontology integration and evaluation is presented in this section.

A. A ONTOLOGY INTEGRATION
Ontology integration refers to the process of establishing and processing mappings between ontology entities to achieve ontology alignment or ontology merging when multiple heterogeneous ontologies are applied to an ontology task [18]. Establishing accurate ontology mapping is a basic task and core component of ontology integration. According to an ontology definition model, studies on ontology mapping can be divided into grammar-based methods, concept instance-based methods, concept definitionbased methods, and concept structure-based methods [19]. Grammar-based methods involve calculating the edit distance of the concept name [20], [21] and calculating the basic distance between two nodes [22]. A typical representative of the concept instance-based methods is the GLUE system of Washington University [23]. Concept definition-based methods calculate the similarity between concepts by using concept definition [24]. Concept structure-based methods consider the hierarchical structures between concepts when mapping, e.g., node relationships, semantic neighbor relationships, etc. Because of the large number of latent semantics in the hierarchical relationship of nodes, this approach has been adopted in many mapping methods; typical representatives include [22] and [24]. In addition, there are other ontology mapping methods such as rule-based ontology mapping methods [25], [26], statistics-based ontology mapping methods [23], [26], etc. Although these ontology mapping methods are diverse, their shortcomings are also obvious. Mapping methods based on various similarity measures, such as those using ''recall'' and ''precision'', are mostly limited to measuring the equivalence relationships between the entities, emphasizing grammar implementation and lacking an accurate description of semantics. Rule-based methods extract the ontology connotations through semantics and lack a relationship consistency test. Statistical methods are prone to computational errors. The introduction of a DL strategy can avoid the above deficiencies. Different researchers [27]- [29] have conducted related studies; however, most applications only involved the same or similar domain ontology (DO) concepts and rarely involved role levels; moreover, even if role matching was mentioned, it was limited to direct matching, and no intermediate concept or role delivery mapping was used for indirect matching.
Based on the above review and analysis, this paper proposes a hybrid ontology mapping method that combines semantic similarity calculation with DL.

B. ONTOLOGY EVALUATION
Ontology evaluation can be defined as ''a technical judgment of the content of the ontology with respect to a frame of reference during every phase and between phases of their life cycle'' [30]. To achieve the best results and high-quality ontology, one needs to choose from the available list of aspects of ontology to be evaluated; the right approach to evaluation; the right mix of criteria to be evaluated; and also the right tools to be used [31].

1) ASPECTS
Aspects include the vocabulary, syntax, structure, semantics, representation and context of the ontology, which are defined according to literature [31]- [33].

2) APPROACHES
The different known methods and techniques can be mainly assigned to four different kinds of approaches: technologybased, quality-attribute based, data-driven and application or task-based evaluation [31], [34]. Technology-based evaluation investigates the syntax, consistency and formal semantics and thereby ensures the correctness and usability of the ontology. Its typical representative is OOPS!, a web-based tool which is accompanied by a catalogue of potential and common pitfalls [35]. However, this approach cannot tell anything about the quality of the content and applicability of the ontology [36], [37]. Quality-based approach offers a quantitative evaluation which relies on a set of predefined metrics that measure individual quality attributes of an ontology. Yet, some of those quality metrics tend to be hard to measure and might need human experts to evaluate [33]. Its typical representatives include, OntoClean methodology, OntoMetric [38], OntoQA [39], etc. Data-driven evaluation approach concentrates on the usability of an ontology considering its future application and has also been the current focus of recent research [40]- [42]. This approach attempts to analyze how adequate an ontology covers the domain but is not applicable to determine the correctness or clarity of the ontology [43], [44]. Application or task-based evaluation approach would typically involve evaluating how effective an ontology is in the context of a specific application [45]. This approach exhibits a limitation: the result obtained from one task may not be useful for another task as each task is different [34], i.e., it is not suited for a general evaluation, because every ontology must be evaluated individually depending on the application context [43].

3) CRITERIA
This kind of evaluation approach is done by humans who try to assess how well the ontology meets a set of predefined criteria, standards, requirements, etc. Reference [46] Various criteria have been proposed in literature to evaluate the quality of ontology [31]- [33], [47]: consistency, completeness, accuracy, conciseness, correctness, computational efficiency, adaptability, clarity.

4) TOOLS
Various tools have been developed to support the task of ontology evaluation, each concerned with different aspects of evaluation. There exist tools for checking the consistency, the structure or modeling mistakes of the ontology [37]. Various available tools include: ODEClean, ODEval, AEON, Eyeball, Moki, XD-Analyzer, OQuaRE, OntoCheck, OntoQA, OntoClean, OntoMetric, ACTiveRank, OOPS!, ODEval, oQual [31]. VOLUME 8, 2020 Based on the above review and analysis, this paper adopts a criteria-based evaluation approach to evaluate the quality of the integrated ontology, including ontology validation and ontology verification.

III. CONSTRUCTION OF THE UAV FCMS SREO AND GEO-ONTOLOGY
Currently, the widely accepted ontology construction guideline is the five criteria proposed by Gruber, i.e., clarity, coherence, extendibility, minimal encoding bias, and minimal ontological commitment [4]. Moreover, there are other supplementary rules for specific operations. The most famous rules are Arpirez's three criteria, i.e., the standardization of concept name, the diversification of concept level, and the minimization of semantic distance [48]. Following the above rules, combined with the engineering application background of this research, this paper uses the TOVE method [49] to guide the ontology construction.

A. ONTOLOGY FORMALIZATION
Ontologies provide interrelations between elements, hierarchy among domain concepts, data structure and the integration of heterogeneous information [50]. The different ontology classes, relationships, constraints and axioms define a common vocabulary to share knowledge [51].
Formally, an ontology can be defined as the tuple: where: C = C C ∪ C I is the set of entities of the ontology. The set C C consists of classes, i.e., concepts that represent entities that describe a set of objects, while the set C I is constituted by instances. H = {kind_of (c 1 , c 2 )|c 1 ∈ C C , c 2 ∈ C C } is the set of taxonomic relationships between the concepts, which define a concept hierarchy and are denoted by ''kind_of (c 1 , c 2 )'', meaning that c 1 is a subclass of c 2 .
. . , c n )|∀i, c i ∈ C I } is the set of relationships between ontology elements and its instances.
is the set of ontology relationships that are neither ''kind_of'' nor ''is_a''. The relationships between concepts mainly have two types: hierarchical relationships and non-hierarchical relationships [52].
is the set of properties of ontology entities and its basic datatype.
A = {condition x ⇒ conclusion v (c 1 , c 2 , . . . , c n )|∀j, c j ∈ C C } is a set of axioms, rules that allow checking the consistency of an ontology and infer new knowledge through some inference mechanism. The term ''condition x '' is given by condition x = {(cond 1 , cond 2 , . . . , cond n )|∀z, cond z ∈ H ∪ I ∪ I ∪ R}

B. UAV FCMS SREO CONSTRUCTION 1) THE UAV FCMS SREO CONSTRUCTION PROCESS
A UAV FCMS SREO has a variety of contents that involve both the UAV FCMS field and the software engineering field; therefore, the knowledge system can be modeled by a knowledge aided design system (KADS) [5]. The knowledge hierarchy in this model is clearly divided, and each layer of knowledge exhibits good maintainability and reusability. Furthermore, to enable the above knowledge model to play a role in knowledge sharing and reuse, it is necessary to integrate relatively independent knowledge layers through the ontology to form a knowledge system. This paper constructs both generalization layer and domain layer ontologies. The UAV FCMS SREO construction process includes the elicitation of domain knowledge; the elicitation of concepts, concept attributes, concept hierarchies and concept relationships; and the use of a formal language to represent these definitions.

2) UAV FCMS SREO CONCEPTS AND RELATIONSHIPS
• UAV FCMS software-related concepts and relationships First, a generalized ontology (GO) is constructed according to the KADS. Figure 1 shows the hierarchy of the concept classes in the GO. The concept class with a '' * '' is a nonterminating concept class, and the rest are all terminating concept classes. Furthermore, a portion of the concept dictionary table and a portion of the GO concept space are shown in Table 1-Part A and Table 2 , respectively.
Second, the UAV FCMS SRE DO is built. The UAV FCMS software is the core part of the UAV FCMS. Figure 2 shows the internal structure and main external interfaces of the UAV FCMS.
Due to the variety of concepts involved, in the concept selection stage, this paper uses the term 'weighting technique' along with equation 2 [53].
A portion of the concept dictionary table is shown in Table 1-Part B.
• The related concepts and relationships of the SREP Definition 1: The SREP refers to the error produced in the software requirement development stage, which occurs repeatedly in a specific error lifetime scenario, spreads in the subsequent design and implementation, and may cause a system (component) to fail to perform the expected function or affect the maintainability of the system. Such errors  are general and common in a specific scenario and can be corrected by various means.
Furthermore, for the sake of simplicity, this paper selects these three concepts as part of the collection of ontology concept classes for the UAV FCMS SREO. The concept dictionary table is shown in Table 1-Part C.
• The concept classes and relationships of the UAV FCMS SREO Because the concepts and relationships associated with the SREP are relatively independent of other concepts and relationships of the UAV FCMS SREO, Figure 3 only shows the unified model language (UML) diagram representations of the concepts and relationships of the UAV FCMS SREO other than the SREP. '' '' represents the inheritance relationship and '' '' represents the relationships other than the inheritance relationship.

C. THE SELECTION OF GEOGRAPHIC INFORMATION METADATA AND DOMAIN ONTOLOGY CONSTRUCTION 1) GEOGRAPHIC INFORMATION METADATA
The use of digital geographic data is intended to simulate and describe the real world for computer analysis and the graphical display of information [13]. In the current digital geographic data domain, the authoritative and available domain concept classification standard and domain system structure include the contents given in literature [12] and [13]. The FGDC and ISO TC/211 assert that metadata contain data on data content, quality, conditions, and other characteristics.
However, the contents residing at the geographic information metadata level are insufficient for actual domain use. This is due to significant differences between the metadata and the ontologies, (1) the metadata mainly focus on the external form features of information resources, whereas an ontology mainly focuses on the inherent content characteristics of the information resources; (2) the metadata focus on the description and positioning of the information resources, while an ontology organizes and manages the knowledge content. More critically, metadata lack semantic description capabilities; therefore, they cannot solve the problem of the semantic heterogeneity of data sets or the description of the implicit relationships between data categories. Thus, it is necessary to establish an ontology layer on the top of the metadata and perform semantic description and ontology reasoning.
In this research, domain experts built a hierarchical concept system by selecting the parts associated with the UAV FCMS SRE considering the geographic environment factors in the field of digital geographic data, and each concept was described by a set of attributes. Because our research focuses on the UAV-related geo-ontology construction, the main role of the UAVs in this case is intelligence collection. According to literature [13], the corresponding ontology was defined as an ''intelligent military domain ontology'' (IMDO). Table 3 outlines a portion of the intelligent military domain-related geographic information metadata.   Table 3 shows that these contents contain rich feature information such as longitude, latitude, height, etc. Moreover, they express not only certain geographic information semantics but also other intelligent military domain-related information. Therefore, the relationships between the standard feature sets of different concepts can be found based on the shared feature attribute sets of concepts of different ontologies. Moreover, the integration between different domain concepts can be realized based on the construction of corresponding conceptual systems and architectures. Section 3 addresses this challenge.

2) CONSTRUCTION OF AN IMDO
The greatest difference between a geo-ontology and a general ontology is that the former possesses not only general attribute characteristics but also spatial characteristics; thus, an IMDO also has such characteristics. The main idea in describing geographic element-related concepts in the IMDO is to divide the described objects into two categories: conceptual attributes and spatial attributes. The conceptual attributes describe the non-spatial ontology attributes in terms of five aspects-matter, form, spatial distribution, function, and rank, while the spatial attributes describe the ontology in terms of three aspects-topological relationship, positional relationship, and directional relationship. The following takes the Yangtze River as an example to describe the semantic features. The OWL code for the formal description of the Yangtze River is shown in Figure 4.

IV. THE INTEGRATION OF THE UAV FCMS SREO AND THE IMDO A. THE HYBRID ONTOLOGY MAPPING METHOD
This paper realizes the integration of the UAV FCMS SREO and the IMDO through ontology mapping. It performs a similarity analysis [14] of the concepts in the above two ontologies. The specific processes include, lexical comparison, structural comparison and relational comparison.
However, the semantic similarity analysis method is limited to the measurement of the equivalent relations between the entities and lacks a precise description of the semantics.  In some cases, the semantic similarity values obtained solely by this method are not accurate. Thus, it is necessary to use DL to detect the matching relationship between the concepts, as well as between the concept and role in different domain ontologies, and realize the matching from one ontology to another. For specific processes, an ontology API is initially used to parse the two ontologies to be integrated, and the concepts and roles are acquired; then, a data dictionary is used to complete the string matching of the concepts and roles; finally, an inference engine performs reasoning to make the concepts and roles in one ontology gradually match the other ontology according to inference rules. The process of the hybrid ontology mapping method is shown in Figure 6.
The activity ''lexical comparison'' performs a lexical comparison between the representative terms of the elements obtained from the two ontologies O 1 and O 2 . This activity takes as input two lists of terms ''list of terms in O 1 '' and ''list of terms in O 2 '' composed by elements of the sets C C , P, R and H of the ontology definition in section III. Then, each term is enriched with its synonyms by consulting a lexical dictionary. Then, for each ontology element, the terms and their synonyms are compared with the correspondent terms of the other list. For the result, two types of values can be obtained: 1 for perfect match and 0 for no match. In addition, we also need to define synonyms in combination with domain features.

2) STRUCTURAL COMPARISON
The activity ''structural comparison'' makes a similarity analysis between the terms in the sets C C of the ontologies O 1 and O 2 . A similarity measure which considers the hierarchical structures in which they are inserted was adapted from the work of Mendes and Girardi [54].
where c i is a class of ontology O 1 ; c j is a class of ontology O 2 ; C iH is the list of super classes of class c i in the hierarchy H; C jH is the list of super classes of class c j in the hierarchy H.

3) RELATIONAL COMPARISON
The activity ''relational comparison'' performs a similarity analysis between the non-taxonomic relationships in the ontologies. Thus, when a lexical similarity is found between two relationships R 1 and R 2 of the ontologies O 1 and O 2 , a weight is assigned to the result of the structural comparison of the concepts related by them. This assignment of weights based on the identification of elements of the set R, increases the value of the similarity measure and makes it more adequate once the comparison uses the whole structure of the ontology realizing a semantic comparison between its elements.

C. DESCRIPTION LOGIC 1) THE FEASIBILITY OF DL USAGE
To realize the integration of a UAV FCMS SREO and an IMDO, it is necessary to clarify the semantic relationships VOLUME 8, 2020 between the ontology concepts. Because the concept level can ignore the extension of concepts, i.e., the instance sets, the semantic relationships of concepts are completely determined by their connotation relationships. The calculation of the connotation relationships between the UAV FCMS SREO concepts and the IMDO concepts involves calculating the concept attribute set and its range. It is a set operation that satisfies the typical set operation syntax and can define the concept connotation relationships as four semantic relations: synonymous relationships (semantic equivalence relations), upper and lower semantic relationships (parent/child concept relationships), semantic intersections and semantic nonintersections. Therefore, the ontology integration can be studied by DL.

2) THE MATCHING METHOD BASED ON THE DL
DL is descended from so-called ''structured inheritance networks'' with the basic components as concepts, roles, and individuals [15] and is widely used as the basis for ontology description languages. The vocabulary consists of concepts, which denote the sets of individuals, and roles, which denote the binary relationships between the individuals. A DL system consists of four parts: a constructor set representing the concepts and relationships, a reasoning mechanism on Tbox/Abox, the Tbox, and the Abox. The Tbox is a set of axioms that describes the structure of a domain, including concept definitions and the inclusion relationships of concepts. It is implemented through a set of statements describing the general attributes of concepts. Connotation axioms are invariant. The Abox is a set of axioms that describes the named individuals. It contains the extended knowledge, including instance assertions and relational assertions. The extended knowledge is often considered to be constantly changing [16]. Usually, a DL system contains at least the following constructors: intersections(∩), unions(∪), negations(-), existential quantifiers(∃), ∃ universal quantifiers(∀∀), bottom concepts (⊥), universal concepts ( ) [15]. Complex concepts and roles can be constructed through simple concepts and relationships [51].
• The definition of a matching relationship Definition 2: For a C (a concept or role) of O i and a D (a concept or role) of O j , iff, an arbitrary individual, satisfying the following five mappings in turn, means that the relationships between the two are an equivalence relation, a subsumption relation, a supersumption relation, an overlapping relation or a disjoint relation respectively, i : where: 5) and (6) are mutually inverse.
• Direct inference The implicit knowledge contained in a DL knowledge base can be made explicit through inferences [15]. The following describes the reasoning rules for the two aspects of ''concept level'' and ''concept vs. role'' [28], [29].

a: CONCEPT LEVEL
From the perspective of Abox, the mapping rules of the relationships between the concept ''X'' of O i and the concept ''Y'' of O j are given, and ''a'' represents an arbitrary individual.
To determine the correspondence between these two ontologies, the analysis of the concept, as well as the range and domain of the role, must be examined. Set Rule 6: Tbox Concept vs. role subsumption, Rule 7: Tbox Concept vs. role overlapping, • Transitive inference A direct inference is limited to the relationships between two concepts or between a concept and a role. In practice, it is often difficult to find a direct mapping relationship between the two; therefore, it is necessary to use a transitive inference through intermediate concepts [16]. As shown in Table 4 , X 1 , X 2 and X 3 denote the concept or role. ∼ = and & denote that the relationship between two concepts is a fuzzy relation (⊆, ⊇, &) or a disjoint relation.
Rule 9: In short, guided by the process of the hybrid ontology mapping method, using a semantic similarity analysis and a DL strategy, the integration of a UAV FCMS SREO and an IMDO is finally realized.

V. CASE STUDY
This paper uses the UAV FCMS SREO and the IMDO as experimental objects to perform an ontology integration and evaluation. The UAV FCMS SREO is mainly based on relevant literature, industry standards, and the development and testing experience of multiple continuous versions of a certain type of UAV FCMS software. Because of the need to consider the geographic environment factor on the UAV FCMS SREO, the hybrid ontology mapping method based on semantic similarity analysis and DL is used for ontology integration. Moreover, the improved FOCA method is used to evaluate the quality of the newly integrated ontology. In addition, the results of engineering applications also illustrate the effectiveness of the method. It should be noted that in this study, only a portion of the UAV FCMS SREO is related to the IMDO.   Figure 5}. The list of equivalent terms shown in Table 5 is obtained based on the conventional lexical comparison method with the domain features.
• The structural comparison The structural comparison activity is performed with the aim of analyzing the similarity between the hierarchies of concepts presented in the ontologies; therefore, equation (3) is used. The intersections in this equation are defined from the equivalences described in Table 5 . The result varies from 0 to 1, depending on how similar is the hierarchical structure between the lists. It is noted that for the concepts of different levels (the generalization layer or domain layer), the similarity values should be calculated in combination with the path of the corresponding level. O 1 has been divided into a generalization layer and a domain layer, while for O 2 , except for ''Thing'', ''System'', ''Agent'' and ''Environment'' belonging to the generalization layer, the remaining values should belong to the domain layer.  (3), Taking the terms ''SpatEn'' and ''Extent'' as an example, O 1 : ImpEn→SpatEn O 2 : DataTypeInfo→Extent Substituting the values into equation (3), = 0.50. Table 6 presents the similarity values between the concepts of the two ontologies.
• The relational comparison The relational comparison is performed with the aim of analyzing the similarity among non-taxonomic relationships of the ontologies. Thus, for each lexical similarity found between terms belonging of the set R, weights are assigned to the values obtained from the activity of structural comparison. Table 7

2) DESCRIPTION LOGIC
It can be seen from Table 7 that the similarity values of some concepts are improved after the weights are added. However, the overall similarity values of the concepts are still low, which is inconsistent with the actual domain situation and does not fully reflect the true semantic information. In addition, there are some concepts with new semantics that require new concepts to be added. Therefore, it is necessary to adopt the DL strategy further. The DL strategy is performed according to the process in Figure 6.  Table 7 from ''A: HeightUnit'' to ''A: Strait'' are each equivalent to the concepts of the second column in Table 7, and the corresponding concepts of the two columns can be merged.  Using equation (21) Therefore, these two concepts can be merged. The list of terms in the integrated ontology O N is shown as Table 8. (only for the concepts and attributes that exist in both ontologies before integration.) VOLUME 8, 2020  Figure 3, a network diagram of concept classes and relationships of the newly integrated ontology is shown in Figure 7.

Based on
The ontology integration process described above is semiautomated. In addition to the automatic reasoning using the inference engine, human participation is also required to delete some concepts or confirm the reservations of certain concepts manually. For example, the concept ''DataType-Info'' in the IMDO does not exist in the newly integrated ontology because from a semantic point of view, this concept is not needed in the new ontology.

B. ONTOLOGY EVALUATION
The development of ontology description languages and tools aids developers in building ontologies according to specific applications. However, due to the complexity of application semantics, ensuring ontology quality remains an important issue. In addition, the widespread use of ontologies has led to an explosive growth in the number of ontologies on the Internet. Ontologies enable reuse, but different ontologies have notable differences in domain coverage, comprehensibility and accuracy. Thus, it is difficult for users to grasp ontology features as a whole and understand their application. Based on the above two points, it is necessary to evaluate ontology quality. According to ontology evaluation results, developers can reconstruct an ontology to optimize its structure, thereby creating high-quality ontologies. Meanwhile, users can also select an optimal ontology between different ontology systems.
This criteria-based evaluation is our approach to measure internal and external semantic structural domains and concept structures in ontologies via our proposed criteria. It consists of ontology validation and ontology verification.

1) ONTOLOGY VALIDATION
• Ontology content evaluation This method checks the content of the ontology based on the following main criteria [31]- [33], [47]: consistency, completeness, accuracy, conciseness, expandability, and sensitiveness. The criteria and their compatibility to UAV FCMS SREO considering geographic environment factors are shown in Table 9.
• Competency questions evaluation The competency questions for determining the scope and designing purposes of UAV FCMS SREO considering geographic environment factors are used here for the evaluation. Answers and justifications are shown in Table 10. Competency questions ensure that the ontology implementation fulfills the scope of UAV FCMS SREO considering geographic environment factors.

2) ONTOLOGY VERIFICATION
• Ontology taxonomy evaluation The taxonomy evaluation method is used for checking the taxonomy of the ontology based on main criteria mentioned in [55]. These criteria and their compatibility to UAV FCMS SREO considering geographic environment factors are shown in Table 11. FOCA is a method that can be used for evaluating the quality of an ontology. FOCA includes determining the type of ontology, a questionnaire to evaluate the components, a framework to follow, and a statistical model that calculates the quality of the ontology. FOCA goes through three verification steps, as shown in Figure 8 [17]. Ontology type verification defines two types of ontology: a domain or task ontology and an application ontology. Questions verification possesses questions to serve the goals. Quality verification calculates the ontology quality.
The FOCA evaluation criteria do not include a quantitative evaluation of ontology cohesion reflecting the close relationship between the ontology concepts. The structure of an ontology is consistent with object-oriented structure and should also meet the principle of ''high cohesion, low coupling''. The higher the ontology cohesion is, the closer relationship between the concepts. Therefore, the ontology cohesion can reflect the degree of ontology modularization to a certain extent. More importantly, because of the ontology integration technology used in this paper, the cohesion of the related concepts in the integrated ontology is also an important indicator reflecting the ontology quality. This paper calculates the cohesion of the parts related to both original ontologies of the newly integrated ontology and adds this indicator to the FOCA to evaluate the ontology quality.

b: ONTOLOGY MODULE AND DIRECTED ACYCLIC GRAPH
An ontology module is a collection of the closely related concepts, relationships, and axioms reflecting a common theme. The ontology module is divided or extracted from an original ontology and is part of the original ontology [56].  The modularization of an ontology helps reduce complexity and enhances comprehensibility, testability, maintainability, and reliability. The module has its own cohesion and can be used independently [57].
Ontology classes are arranged in a hierarchy from the general (high in the hierarchy) to the specific (low in the hierarchy). Despite the hierarchical organisation, most ontologies are not simple trees. Rather, they are structured as directed acyclic graphs (DAGs). This is because it is possible for classes to have multiple parents in the classification hierarchy, and furthermore ontologies include additional types of relationships between entities other than hierarchical classification (which itself is represented by is_a relations). All relations are directed and care must be taken by the ontology editors to ensure that the overall structure of the ontology does not contain cycles, as illustrated in Figure 9 [58].

c: THE EVALUATION METRICS OF COHESION
This paper uses the following evaluation metrics of cohesion [59]: the ontology module cohesion ''Coh'', the original ontology cohesion ''AOC'', the leaf node average VOLUME 8, 2020   The calculation equation of AOC is, where n is the number of modules partitioned by the original ontology, and Coh(M i ) is the cohesion of the ontology module M i . ADIT-LN represents the depth of conceptual hierarchy in the ontology and depicts the degree of richness and refinement of the concepts. The calculation equation is, where DOI i represents the inheritance depth of a path i from its root node to a leaf node in a DAG, and the inheritance depth refers to the total number of edges of the path i from its root node to a leaf node in a DAG. n = TNOP, i.e., the total number of different paths from the root node to the leaf nodes in a DAG. TCOO is calculated as follows, where α + β = 1, O represents the original ontology. The values of AOC and ADIT-LN are relatively high, indicating that the relationship between the original ontology concepts is closely connected; moreover, the concept level of the original ontology is relatively deep, and the concepts are rich.

d: THE CALCULATION OF ONTOLOGY COHESION
In this paper, the ontology cohesion is calculated for a portion of the integrated ontology related to both original ontologies (also an ontology module). It should be noted that because ''Agent'' and the related concepts account for a relatively small amount, the influence on cohesion is not significant. Therefore, this paper only calculates the cohesion of the ''Environment'' module. ''Environment'' is the apex of this ontology module, as shown in Figure 10.
According to the guidelines for prioritizing the protection of the hierarchical relationship, M1 and M2 are obtained by the modularization using the module partitioning tool SWOOP [60], i.e., the hierarchical relationship between concepts is not destroyed in modularization process.
Using equation (32)  Let αα = 0.60 and ββ = 0.40; using equation (35), TCOO(O) = 0.767. This is the comprehensive cohesion of the ontology module with ''Environment'' as its apex in the integrated ontology. Referring to the results of the case study section in [59], the comprehensive cohesion of the main integrated part of the newly integrated ontology in this paper is slightly low. It should be noted that the AOC value is low, indicating that the relationship between the concepts in the ontology is not very close. Furthermore, the ADIT-LN value is not high, indicating that the concept hierarchy is not sufficiently deep. The main reason for the above results is that the hierarchical structure of the conceptual classes in the ontology is not sufficiently complete; meanwhile, there are limited connections between the concepts other than the hierarchical relationships. Ontology type verification FOCA defines two types of ontology, a domain or task ontology and an application ontology. The UAV FCMS SREO considering geographic environment factors is a DO (type 1); therefore, a type 1 ontology should answer Q5 instead of Q4 for Goal 2 shown in Table 12.

Questions verification
When a cohesion metric is added, it needs to answer the 13 questions in Table 12 (should answer Q5 instead of Q4). These answers should then be scored by the evaluator. The set of questions corresponding to Goal 2 is expanded by adding a question of ontology cohesion metric, i.e., ''Was the ontology cohesion metric value acquired?''. The scores refer to the experimental data in [59]. The cohesion metric values and corresponding scores are shown in Table 13. These 13 questions serve five goals. The goal/question/metric (GQM) approach for the improved FOCA is shown in Table 12.

Quality verification
Ontology quality can be calculated in two ways: total quality and partial quality. This paper uses the total quality verification because most goals are considered in the evaluation. Total quality verification is calculated using beta regression models, proposed by Ferrari [61], and shown in (36), as shown at the bottom of next page.
• Cov S is the mean of the grades from Goal 1.
• Cov C is the mean of the grades from Goal 2.
• Cov R is the mean of the grades from Goal 3.
• Cov Cp is the mean of the grades from Goal 4.
• LExp is the variable for evaluator experience, with 1 being very experienced and 0 being not experienced at all.
• Nl is 1 only if some Goal is impossible for the evaluator to answer all the questions.
• Sb = 1, Co = 1, Re = 1, Cp = 1, because the total quality considers all the roles. VOLUME 8, 2020   The total quality of the ontology is 0.9725, which is near to 1. This shows the high quality of the UAV FCMS SREO considering geographic environment factors.

C. ENGINEERING APPLICATIONS
The ontology proposed in this paper has already been applied in engineering, i.e., the requirement elicitation of a certain type of UAV FCMS software has been carried out based on the ontology this paper proposes. To illustrate the effectiveness of this method, further verification is necessary using a comparative experiment (a software requirement inspection). The research selects two continuous versions of the UAV FCMS software and adopts a conventional method for the requirement elicitation of a version 3.3.x; after a defined period, the requirements of a version 3.3.(x + 1) are elicited based on the ontology this paper proposes. Table 14 records the detected SREP error-manifestations and number distributions of these two SRSs by a requirement inspection. The same group of inspectors is used to conduct the comparative experiment. SRS I is developed based on the conventional method, and SRS II is developed based on the ontology proposed in this paper.
The results that the total number of errors in SRS I is much higher than in SRS II. In addition, the severity of the errors detected in SRS I is higher, and they occur in more significant error types such as functional errors, interface errors, safetŷ  errors, and environmental errors. The direct cause of the above results can be initially identified as SRS II using the ontology-based method; SRS I uses the conventional method.
Intuitively, because the ontology is a complete set of domain knowledge, the quality information of SRS can be obtained indirectly by considering the correspondence between the SRS and ontology element. This paper evaluated the quality of SRS based on the metrics in [62]. The quality metrics results of the SRS I and SRS II are shown in Table 15.
From the results, the difference between SRS I and SRS II in ''Correctness'' is more obvious. An ontology is a semantic basis for building a specific problem domain. Ideally, all requirements items should be able to find the corresponding elements in the ontology. (the number of items that can be mapped to the ontology / the total number of requirements items) can reflect the proportion of the mapped elements. The higher the ratio is, the higher the SRS quality. This ratio of SRS I to SRS II is significantly lower. This shows that some of the requirements items of SRS I are not included in the ontology library, implying nonconformity with the actual application. This fact also explains the results of the requirement inspection in Table 14. Therefore, the requirement knowledge ontology has a major impact on the entire requirement development process. In general, the quality of the SRS obtained based on the proposed ontology is higher than the quality of the SRS obtained based on the conventional method. Therefore, the ontology elements, i.e., the knowledge elements, should be fully integrated in the early stage of the requirement development process.

VI. CONCLUSION
This paper focused on the problems of ambiguity, heterogeneity, and incompleteness in a UAV FCMS SRE, especially geographic environment-related factors, to use an ontology to solve the above problems. By constructing a UAV FCMS SREO and an IMDO and integrating these two ontologies, a UAV FCMS SREO considering geographic environment factors was obtained. A hybrid ontology mapping method was adopted to analyze the ontology concepts. Based on a traditional similarity calculation, a DL strategy was used to detect the matching relationships between different domain ontologies through deductive reasoning, realize the mapping between two ontologies, and finally complete the ontology integration. This method avoided the shortcomings of the similarity calculation method, which was limited to measuring the equivalence relation between the entities, merely emphasizing the grammar implementation, and lacking an accurate description of the semantics. The ontology evaluation results showed the higher quality of the integrated ontology. Moreover, the engineering applications showed that the SRE activities and the generated SRS based on the proposed ontology enabled a notable increase in quality.
However, the results in this paper are still insufficient, and the new cohesion index in the improved FOCA is not very satisfactory. This suggests that the relationship between the concepts in the UAV FCMS SREO considering geographic environment factors is not very close; furthermore, the concept hierarchy is not sufficiently deep. The main reason for the above results is that the concept hierarchy in the ontology is not sufficiently complete, and the relationships between the concepts are limited except for the hierarchical relationships. Therefore, it is necessary to further improve the ontology, enrich and refine the ontology concepts, and fully explore the implicit relationships between the concepts.