1. bookAHEAD OF PRINT
Journal Details
License
Format
Journal
eISSN
2444-8656
First Published
01 Jan 2016
Publication timeframe
2 times per year
Languages
English
Open Access

Animation character recognition and character intelligence analysis based on semantic ontology and Poisson equation

Published Online: 15 Jul 2022
Volume & Issue: AHEAD OF PRINT
Page range: -
Received: 15 Mar 2022
Accepted: 20 May 2022
Journal Details
License
Format
Journal
eISSN
2444-8656
First Published
01 Jan 2016
Publication timeframe
2 times per year
Languages
English
Introduction

The most popular concept of artificial intelligence was first proposed by computer experts in 1956. Although it has experienced several cold winters in the following decades, with the rapid development of data resources, computing power and core algorithms, artificial intelligence has evolved from computational intelligence to higher-level perceptual and cognitive intelligence. In order to occupy a place in the market, animation related manufacturers continue to launch new animation works and applications in their own products in order to retain users and attract users. The integration of artificial intelligence technology into products will be a very novel and interesting application [1]. The intelligent analysis of animation characters in this paper is such an application. It analyzes the animation characters contained in the image in the form of target detection, which can help users easily understand strange animation characters, arouse users' interest in animation characters and related works, and make animation products more entertaining and attractive. As we all know, the necessary premise of a large number of advanced computer vision tasks is target detection, such as scene content understanding, event activity recognition and so on. Target detection is also widely used in many practical tasks, such as robot navigation, intelligent video surveillance, content-based image retrieval and augmented reality. However, current target detection also has some niche areas that are not involved, such as animation characters, and its detection performance in these areas is unknown. Although the research related to target detection has increased day by day in recent years, new research results and practical applications have been published [2]. However, the detection accuracy of mainstream methods still needs to be improved, and can not be applied to practical and general target detection tasks. Therefore, animation character recognition and character intelligence analysis based on semantic ontology and Poisson equation has certain exploration and scientific research significance. The intelligent recognition of animation characters discussed by Lembo D and others is mainly based on target detection method. Target detection is one of the basic tasks in the field of computer vision. It needs to find the specific location of the specified target and accurately predict the category of the target [3]. Among them, category prediction is actually a classification task. Its goal is to select the category label that is most consistent with the target object from all labels in the training set; The purpose of location is to accurately give the position of the object in the image. The position is generally represented by the bounding box. Finally, through classification and target location, the algorithm can automatically locate multiple target objects in a map. Alkhutov Y A and others said that the basic problem in target detection is classification, and the positioning problem can be realized by a variety of means. For example, in the traditional target detection algorithm, the sliding window algorithm is often used to intensively extract the window features in the image, and then the target category prediction and location are realized by classifying a large number of windows [4]. From the existing research, for animation character model recognition, the consistency of recognition results in hierarchy can be effectively guaranteed through a priori knowledge. However, the existing research has not discussed how to build a standardized knowledge structure to define various parameters required for recognition, and these parameters have a very important impact on the quality of recognition. Therefore, based on the existing research, Chen J and others proposed to build a recognition book to form an ontology knowledge definition suitable for animation character recognition. The definition of recognition ontology is proposed for the animation character model, so that users can customize the recognition hierarchy, shape features, optimization algorithm and recognition boundary optimization parameters to obtain the optimal recognition effect [5]. On the problem of recognition boundary consistency, aiming at the characteristic that the recognition boundary of three-dimensional model has equal circumference under different attitudes, the recognition boundary optimization is proposed by Poisson equation, so as to make the recognition boundary more consistent.

Method
Animation character recognition based on Semantic Ontology

Considering the animated character model with obvious structural characteristics, the sum of geodesic distances of its external structural points has extreme characteristics. Therefore, the external structural points can be well detected through the protrusion function, and these stable feature points can be used to drive the first layer of coarse recognition [6]. The protrusion function is the integral of the geodesic distance along the surface of the three-dimensional model, which is defined as equation 1: f(pi)=j=0Ng(pi,pj) f\left({{p_i}} \right) = \sum\limits_{j = 0}^N {g\left({{p_i},{p_j}} \right)}

Where: pi and pj are model vertices, I and j are index numbers of model vertices, g (pi, pj) is geodesic distance between two points, and N is the number of surface vertices of 3D model. For the vertex on the mesh, if its protrusion function value is greater than the adjacent vertex φi, it indicates that it is an external feature point. Find a series of local maximum points li, and the set of local characteristic points is L, as defined in equation 2: L={liV,1iNL} L = \left\{{{l_i} \in V,1 \le i \le {N_L}} \right\}

Where: NL is the number of local maximum points, and V is the set of model vertices. Protrusion function is a feature point detection function which is very sensitive to geodesic distance, so it will lead to over detection. Here, we need to combine the number of sub parts defined in the recognition ontology and use K − means clustering to obtain NE − 1 clustering sets (the central point set does not have prominent points, and NE is the number of external feature points of the model) [7]. For the case that there are multiple parts in each region, select the vertex lm with the largest protrusion function value in the region as the initial point of clustering, and the definition of external feature points is as follows: ei=li {e_i} = {l_i}

Where: ei is the detected external feature point. So far, the set of feature points outside the model is defined as E = {eiV, 0 < i < NE}. Because each sub region of the growth model represents a different feature, each sub region is identified. The smaller the value of the protrusion function, the greater the probability that the vertex belongs to the rigid part of the model. Therefore, the priority queue is constructed according to the protrusion function value of the model vertices, and then the vertices are expanded until all the protrusion sub parts are divided into a series of unconnected regions, and the first layer rough recognition is completed [8]. In the implementation of the algorithm, in order to avoid the high complexity of extracting the rigid body part by multi-dimensional calibration, firstly construct the geodesic path between the external feature points, and then track the vertices on the geodesic path until the extracted rigid body region has effectively separated different sub parts, and the algorithm ends. Kuhn munkres algorithm is a classical method to calculate the optimal matching of bipartite graphs. This algorithm is used to calculate the similarity of two sets of concept sets of services. Gm,n = (Om, In, E) is a bipartite graph, Om and In are the set of vertices of the graph, E is the set of edges, where, Om = (o1, o2,…, om), In = (i1, i2,…, in), E = (e11, e12,…, enm), SPPi, j (oi, ij) is the weight of any edge, that is, eij = SPPij (oi, ij). Taking the bipartite graph Gm,n = (Om, In, E) as the input of the algorithm, the specific algorithm is described as follows:

Give the initial label l(oi)=maxjSPPij(oi,ij) l\left({{o_i}} \right) = \mathop {\max} \limits_j \, {SPP}_{ij}\left({{o_i},{i_j}} \right) , l (ij) = 0, where i, j = 1,2,…, t, t = max (n, m);

Use Hungarian algorithm to solve the complete matching M in edge sets El = {(oi, ij)|l(oi) + l(ij) = SPPij(oi, ij)}, GI = (Om, In, El) and Gl;

If M is the complete match in Gl, then M is the optimal match of G. the calculation is over, otherwise proceed to the next step;

Find the unsaturated point o0 of m in Om, let A → {oo}, Bϕ, A and B are two sets;

If PGl (A) = B, turn to step (9), otherwise proceed to the next step, where PGl (A) ⊆ In is the set of nodes adjacent to the node in A;

Find a node i iPGl (A) − B;

If i is the saturation point of M, let AA ∪ {z}, BB ∪ {i}, find out the matching point z of i and turn to step (5), otherwise proceed to the next step;

There is an augmented path R from o0 to i, let MME (R), turn to step (3);

Calculate the value of α according to the following formula: α=minoiAijNGl(A){l(oi)+l(ij)SPPij(oi,ij)} \alpha = \mathop {\min}\limits_{\matrix{{{o_i} \in A} \hfill \cr {{i_j} \notin {N_{{G_l}}}\left(A \right)} \hfill \cr}} \left\{{l\left({{o_i}} \right) + l\left({{i_j}} \right) - {SPP}_{ij}\left({{o_i},{i_j}} \right)} \right\} , modify label: l(v)={l(v)αifvAl(v)+αifvBl(v)other {l^{'}}\left(v \right) = \left\{{\matrix{{l\left(v \right) - \alpha} \hfill & {ifv \in A} \hfill \cr {l\left(v \right) + \alpha} \hfill & {ifv \in B} \hfill \cr {l\left(v \right)} \hfill & {other} \hfill \cr}} \right. Find El′, and Gl′ according to l.

ll, GlGl′, turn to step (6).

Poisson equation

Assuming that ϕij is the unfolding phase on the two-dimensional discrete point (i, j) and φij is the measured principal value phase, there is formula 5: φij=ϕij+n(i,j)+2πk(i,j) {\varphi _{ij}} = {\phi _{ij}} + n\left({i,j} \right) + 2\pi k\left({i,j} \right) k is an integer, −πφijπ. n(i, j) represents the noise in the measurement, and k (i, j) is an integer value function. The task of phase unwrapping is to estimate the appropriate k (i, j) from the principal value phase φij and remove the noise n (i, j), so as to obtain the real continuous phase field φij. Theoretically, the unwrapping phase difference should be equal to the wrapping phase difference. However, due to the existence of singular points, it is impossible to obtain the correct k (i, j) value in the affected region of the interferogram when using equation 5 to unfold the phase [9]. In the process of phase unwrapping, the n (i, j) of the latter point is always calculated on the basis of the previous point. When a singular point is encountered, the phase unwrapping error will be superimposed on the subsequent points, making the error expand along the unwrapping direction, resulting in error bands, that is, the phenomenon of “skip sequence” or “pull line”, which makes there will be a difference between the unwrapping phase difference and the wrapping phase difference. The principle of this algorithm is as follows: find the least squares solution of the difference between the expanded phase difference between adjacent pixels and the wrapped phase difference between adjacent pixels, and obtain the expanded phase in the sense of least squares [10]. First, define the wrapping operator W, as shown in equation 6: W(ϕij)=φij W\left({{\phi _{ij}}} \right) = {\varphi _{ij}}

Define the package phase difference, as shown in equations 7 and 8: Δijx=φ(i+1)jφij \Delta _{ij}^x = {\varphi _{{{\left({i + 1} \right)}_j}}} - {\varphi _{ij}} Δijy=φi(j+1)φij \Delta _{ij}^y = {\varphi _{i\left({j + 1} \right)}} - {\varphi _{ij}}

The superscript x and Y refer to the phase difference of row pixels and column pixels respectively.

Make the least squares on the M × N rectangular pixel grid, as shown in equation 9: S=i=0M2j=0N1(ϕ(i+1)jϕijΔijx)2+i=0M1j=0N2(ϕi(j+1)ϕijΔijy)2+ S = \sum\limits_{i = 0}^{M - 2} {\sum\limits_{j = 0}^{N - 1} {{{\left({{\phi _{\left({i + 1} \right)j}} - {\phi _{ij}} - \Delta _{ij}^x} \right)}^2} +}} \sum\limits_{i = 0}^{M - 1} {\sum\limits_{j = 0}^{N - 2} {{{\left({{\phi _{i\left({j + 1} \right)}} - {\phi _{ij}} - \Delta _{ij}^y} \right)}^2} +}}

The solution of phase unwrapping can be obtained by solving the solution φij in the sense of S least squares.

The following matrix formula gives the normal equation of the above least squares matrix, namely discrete cosine transform (DCT) positive transform, as shown in equation 10: ϕ(i+1)j+ϕ(i1)j+ϕi(j+1)+ϕi(j1)4ϕij=ΔijxΔ(i1)jx+ΔijyΔi(j1)y {\phi _{\left({i + 1} \right)j}} + {\phi _{\left({i - 1} \right)j}} + {\phi _{i\left({j + 1} \right)}} + {\phi _{i\left({j - 1} \right)}} - 4{\phi _{ij}} = \Delta _{ij}^x - \Delta _{\left({i - 1} \right)j}^x + \Delta _{ij}^y - \Delta _{i\left({j - 1} \right)}^y

The formula is the relationship between package phase difference and unfolding phase difference. By making a simple identity transformation, equation 11 is obtained: ϕ(i+1)j2ϕij+ϕ(i1)j+ϕi(j+1)2ϕij+ϕi(j1)=ρij \left\lfloor {{\phi _{\left({i + 1} \right)j}} - 2{\phi _{ij}} + {\phi _{\left({i - 1} \right)j}}} \right\rfloor + \left\lfloor {{\phi _{i\left({j + 1} \right)}} - 2{\phi _{ij}} + {\phi _{i\left({j - 1} \right)}}} \right\rfloor = {\rho _{ij}}

Where: ρij=[ΔijxΔ(i1)jx]+[ΔijyΔi(j1)y] {\rho _{ij}} = \left[ {\Delta _{ij}^x - \Delta _{\left({i - 1} \right)j}^x} \right] + {\left[ {\Delta _{ij}^y - \Delta _{i\left({j - 1} \right)}^y} \right]_ \circ}

In fact, equation 11 is a discrete Poisson equation on a M × N rectangular grid, as shown in equation 12: Δ2Δx2ϕ(x,y)+Δ2Δy2ϕ(x,y)=ρ(x,y) {{{\Delta ^2}} \over {\Delta {x^2}}}\phi \left({x,y} \right) + {{{\Delta ^2}} \over {\Delta {y^2}}}\phi \left({x,y} \right) = \rho \left({x,y} \right)

The above formula is valid for all rectangular grid points i = 0,…, M − 1, j = 0,…, N − 1 and is used to calculate that the phase difference of ρij is non-zero only in the network area [11]. This limitation is derived from the least square formula. Therefore, the Neumann boundary conditions of the discrete Poisson equation can be obtained directly, such as equations 13 and 14: Δ(1)jx=0,Δ(M1)jx=0,j=0,,N1 \Delta _{\left({- 1} \right)j}^x = 0,\,\Delta _{\left({M - 1} \right)j}^x = 0,\,j\, = \,0, \ldots,\,N - 1 Δ0(1)y=0,Δ(N1)jx=0,i=0,,M1 \Delta _{0\left({- 1} \right)}^y = 0,\,\Delta _{\left({N - 1} \right)j}^x = 0,\,i\, = \,0, \ldots,\,M - 1

It is concluded that the least square solution of the difference between the unwrapped phase difference and the wrapped phase difference is mathematically equal to the solution of the discrete Poisson equation with Neumann boundary condition on M × N rectangular grid.

Results and analysis
Experimental analysis

This paper enhances the original animation image data. Data enhancement can prevent the over fitting of training to a certain extent, and plays an important role in the final recognition ability and generalization ability of the trained model. Common data enhancement methods include: horizontal or vertical flipping, random clipping, changing color contrast or brightness, affine or rotational transformation, adding noise and so on. There are different data enhancement methods for different tasks [12]. In order to verify the effect of data enhancement on animation character detection, we use the basic algorithm target detection to test the original animation image data and the enhanced animation image data respectively.

As shown in Figure 1, the abscissa represents the number of iterations in the algorithm training process, and the ordinate represents the result value of the verification set in the training process. The red curve indicates the change of the verification set with the increase of the number of iterations when using the original data as the training set. The green curve represents the change of the verification set with the increase of the number of iterations when using the enhanced data as the training set [13]. It can be clearly seen from Figure 1 that although the value of the red curve is slightly higher than that of the green curve before about 12500 iterations, the value growth of the red curve tends to be flat after that, while the value of the green curve continues to maintain the growth trend and soon exceeds that of the red curve. Finally, both tend to be stable. Finally, the value of the model trained with the original data is stable at about 0.65, while the value of the model trained with the enhanced data is stable at about 0.72, with an increase of 0.07. It can be seen that data enhancement plays a certain role in improving the performance of the algorithm.

Figure 1

Animation image data enhancement curve

As shown in Figure 2, the abscissa is the number of iterations in the algorithm training process, and the ordinate is the value of the loss function in the training process. The red curve indicates that the original data is used as the training set for training, and the value of the loss function changes with the increase of the number of iterations. The green curve represents the change of the value of the loss function with the increase of the number of iterations when using the enhanced data as the training set. It can be clearly seen from the figure that compared with the original data, when using the enhanced data as the training set for training, the convergence curve is more stable and the loss function value of final convergence is lower [14].

Figure 2

Graph of animation image data enhancement loss

Poisson equation experiment and results

In this algorithm, a discrete Poisson equation with Neumann boundary condition is formed for the singular points in the interferogram, and the least square solution is obtained to obtain the correct unfolding phase. Next, the phase unwrapping algorithm in this paper is verified by experiments. The experimental sample is RA 0.094m standard roughness sample block, and the image size is 512 pixels × 512 pixels. Figure 3 is a two-dimensional outline of the phase jump line.

Figure 3

Two dimensional contour of phase jump line

The principle of topography measurement is expressed in the integral form of phase diagram along the x-axis, so the phase unwrapping is also carried out along the x-axis. Firstly, the singular point region is judged. It is assumed that in the interferogram, there is at least one horizontal line that can be correctly wrapped by the phase unwrapping algorithm of equation 5. On this line, all pixels are “good points”. Based on this line, if any square four pixel point formed by the adjacent line and the reference line meets the condition that the sum of the wrapping phase difference is zero, it is correct to unwrap the point on the adjacent line, and the line can be used as the reference line to continue the unwrapping of the next adjacent line. If any square pixel formed by the reference line and the adjacent straight line does not meet the condition that the sum of the wrapped phase difference is zero, it is considered that there are singular points on the connecting line of the pixel. At this time, the phase unwrapping algorithm based on the solution of discrete Poisson equation in the previous section is applied to solve the unwrapping phase in the sense of least squares [15]. In order to reduce the amount of calculation, a small rectangular area with a size of 100 pixels×100 pixels is set at the jump contour of singular points, and the phase unwrapping is carried out according to the programming calculation of the above phase unwrapping algorithm steps. In order to further verify the performance of the algorithm, the recognition quality of this method and the traditional algorithm is compared. The recognition quality mainly depends on two aspects: 1) whether the recognition results are globally consistent in semantic structure; 2) Whether the recognition boundary of the model sequence is consistent. Let the model recognition result of frame X be equation 15: Sx={Fx,Cx} {S_x} = \left\{{{F_x},\,{C_x}} \right\}

Where: Fx is the face index and Cx is the identification index corresponding to each face piece. Define the recognition sequence consistency similarity YF, as shown in equation 16 (assuming that the two models have the same number of points and planes, and the point and plane indexes of the sub parts of any two frame models are also the same): YF(Sx,Sy)=1GTCxCyGTFx {Y_F}\left({{S_x},{S_y}} \right) = 1 - {{{G_T}\left\| {{C_x} - {C_y}} \right\|} \over {{G_T}\left\| {{F_x}} \right\|}}

Where: CxCy represents the difference between the recognition results of the two models, and GT is used to count the number of elements. The identification consistency between the two models can be well measured by equation 16. The larger the YF value, the higher the recognition consistency between the two. Calculate the similarity between the model of M frame and the reference recognition result Sf, and take the mean value of M similarity as the consistency evaluation index ξ of this kind of model, as shown in equation 17: ξ=1Mi=1YF(Sf,Si) \xi = {1 \over M}\sum\limits_{i = 1} {{Y_F}\left({{S_f},{S_i}} \right)}

The traditional algorithm only uses the normal angle between patches as the optimization constraint, so it is difficult to ensure that the recognition boundary converges to the joint position. This method adopts equal perimeter line extraction, and the recognition boundary is not only very smooth, but also has better consistency. In addition, the traditional algorithm uses human proportion as semantic knowledge for semantic label recognition. This method has great limitations and can only be applied to human models. This method constructs a semantic tag recognizer, which only needs to update the training data to get the hierarchical consistency recognition of different models. In conclusion, this method has better universality and applicability.

Conclusion

With the advent and popularity of the era of artificial intelligence, more and more industries begin to try to use artificial intelligence to solve problems or combine artificial intelligence to improve the quality of their products and the attraction to users, so as not to be eliminated in the increasingly competitive market. It is understood that at present, there are few successful cases of the combination of animation products and artificial intelligence elements, and this field is still a blue ocean. Therefore, this paper attempts to apply artificial intelligence to animation images, and realize the intelligent analysis of animation characters by means of target detection based on semantic ontology and Poisson equation. By using ontology to describe role recognition, it can make it semantic. This paper uses the improved Poisson algorithm to solve the problem of intelligent analysis. Finally, based on this correlation degree, the appropriate role recognition and intelligent analysis methods are given, and finally an application service for mobile client is built. Through experiments, it can be found that selecting the optimal parameters through ontology can significantly improve the recognition quality.

Figure 1

Animation image data enhancement curve
Animation image data enhancement curve

Figure 2

Graph of animation image data enhancement loss
Graph of animation image data enhancement loss

Figure 3

Two dimensional contour of phase jump line
Two dimensional contour of phase jump line

Yu B H, Wang H, Dong X P, et al. Design and implementation of a semantic gateway based on SSN ontology[J]. Procedia Computer Science, 2021, 183(6):432–439. YuB H WangH DongX P Design and implementation of a semantic gateway based on SSN ontology[J] Procedia Computer Science 2021 183 6 432 439 10.1016/j.procs.2021.02.081 Search in Google Scholar

Zhang X, Hu Z, Wang M. An adaptive interpolation element free Galerkin method based on a posteriori error estimation of FEM for Poisson equation[J]. Engineering Analysis with Boundary Elements, 2021, 130(1):186–195. ZhangX HuZ WangM An adaptive interpolation element free Galerkin method based on a posteriori error estimation of FEM for Poisson equation[J] Engineering Analysis with Boundary Elements 2021 130 1 186 195 10.1016/j.enganabound.2021.05.020 Search in Google Scholar

Lembo D, Scafoglieri F M. Ontology-based Document Spanning Systems for Information Extraction[J]. International Journal of Semantic Computing, 2020, 14(1):3–26. LemboD ScafoglieriF M Ontology-based Document Spanning Systems for Information Extraction[J] International Journal of Semantic Computing 2020 14 1 3 26 10.1142/S1793351X20400012 Search in Google Scholar

Alkhutov Y A, Chechkin G A. Increased Integrability of the Gradient of the Solution to the Zaremba Problem for the Poisson Equation[J]. Doklady Mathematics, 2021, 103(2):69–71. AlkhutovY A ChechkinG A Increased Integrability of the Gradient of the Solution to the Zaremba Problem for the Poisson Equation[J] Doklady Mathematics 2021 103 2 69 71 10.1134/S1064562421020022 Search in Google Scholar

Chen J, Ou S. Research on the construction of the semantic model for Chinese ancient architectures based on architectural narratives[J]. The Electronic Library, 2020, 38(4):769–784. ChenJ OuS Research on the construction of the semantic model for Chinese ancient architectures based on architectural narratives[J] The Electronic Library 2020 38 4 769 784 10.1108/EL-02-2020-0039 Search in Google Scholar

Gao F C, Lai M J. A New H2 Regularity Condition of the Solution to Dirichlet Problem of the Poisson Equation and Its Applications[J]. Acta Mathematica Sinica, English Series, 2020, 36(1):21–39. GaoF C LaiM J A New H2 Regularity Condition of the Solution to Dirichlet Problem of the Poisson Equation and Its Applications[J] Acta Mathematica Sinica, English Series 2020 36 1 21 39 10.1007/s10114-019-8015-3788826533603567 Search in Google Scholar

Kanna M, Kumar R P, Nandappa S, et al. On Solutions of Fractional order Telegraph Partial Differential Equation by Crank-Nicholson Finite Difference Method[J]. Applied Mathematics and Nonlinear Sciences, 2020, 5(2):85–98. KannaM KumarR P NandappaS On Solutions of Fractional order Telegraph Partial Differential Equation by Crank-Nicholson Finite Difference Method[J] Applied Mathematics and Nonlinear Sciences 2020 5 2 85 98 10.2478/amns.2020.2.00017 Search in Google Scholar

Martino B D, Branco D, Cante L C, et al. Semantic and knowledge based support to business model evaluation to stimulate green behaviour of electric vehicles' drivers and energy prosumers[J]. Journal of Ambient Intelligence and Humanized Computing, 2021(4):1–23. MartinoB D BrancoD CanteL C Semantic and knowledge based support to business model evaluation to stimulate green behaviour of electric vehicles' drivers and energy prosumers[J] Journal of Ambient Intelligence and Humanized Computing 2021 4 1 23 10.1007/s12652-021-03243-4 Search in Google Scholar

Hussain N, HH Wang, Buckingham C D, et al. Software Agent-Centric Semantic Social Network for Cyber-Physical Interaction and Collaboration[J]. International Journal of Software Engineering and Knowledge Engineering, 2020, 30(6):859–893. HussainN HHWang BuckinghamC D Software Agent-Centric Semantic Social Network for Cyber-Physical Interaction and Collaboration[J] International Journal of Software Engineering and Knowledge Engineering 2020 30 6 859 893 10.1142/S0218194020400100 Search in Google Scholar

Safaeipour H, Zarandi M, Bastani S. Mapping Crisp Structural Semantic Similarity Measures to Fuzzy Context: A Generic Approach[J]. International Journal of Fuzzy Systems, 2020, 22(4):1224–1242. SafaeipourH ZarandiM BastaniS Mapping Crisp Structural Semantic Similarity Measures to Fuzzy Context: A Generic Approach[J] International Journal of Fuzzy Systems 2020 22 4 1224 1242 10.1007/s40815-020-00833-w Search in Google Scholar

Nkemzi B, Jung M. The Fourier-finite-element method for Poisson's equation in three-dimensional axisymmetric domains with edges: Computing the edge flux intensity functions[J]. Journal of Numerical Mathematics, 2020, 28(2):75–98. NkemziB JungM The Fourier-finite-element method for Poisson's equation in three-dimensional axisymmetric domains with edges: Computing the edge flux intensity functions[J] Journal of Numerical Mathematics 2020 28 2 75 98 10.1515/jnma-2019-0002 Search in Google Scholar

Alami N, Mallahi M E, Amakdouf H, et al. Hybrid method for text summarization based on statistical and semantic treatment[J]. Multimedia Tools and Applications, 2021, 80(13):19567–19600. AlamiN MallahiM E AmakdoufH Hybrid method for text summarization based on statistical and semantic treatment[J] Multimedia Tools and Applications 2021 80 13 19567 19600 10.1007/s11042-021-10613-9 Search in Google Scholar

Kabra S, Nagar H, Nisar K S, et al. The Marichev-Saigo-Maeda Fractional Calculus Operators Pertaining to the Generalized K-Struve Function[J]. Applied Mathematics and Nonlinear Sciences, 2020, 5(2):593–602. KabraS NagarH NisarK S The Marichev-Saigo-Maeda Fractional Calculus Operators Pertaining to the Generalized K-Struve Function[J] Applied Mathematics and Nonlinear Sciences 2020 5 2 593 602 10.2478/amns.2020.2.00064 Search in Google Scholar

Chang, M., D'Aniello, G., Gaeta, M., Orciuoli, F., & Simonelli, C. (2020). Building ontology-driven tutoring models for intelligent tutoring systems using data mining. IEEE Access, 8(1), 48151–48162. ChangM. D'AnielloG. GaetaM. OrciuoliF. SimonelliC. 2020 Building ontology-driven tutoring models for intelligent tutoring systems using data mining IEEE Access 8 1 48151 48162 10.1109/ACCESS.2020.2979281 Search in Google Scholar

Nath A G, Udmale S S, Singh S K. Role of artificial intelligence in rotor fault diagnosis: a comprehensive review[J]. Artificial Intelligence Review, 2020(45):1–60. NathA G UdmaleS S SinghS K Role of artificial intelligence in rotor fault diagnosis: a comprehensive review[J] Artificial Intelligence Review 2020 45 1 60 10.1007/s10462-020-09910-w Search in Google Scholar

Recommended articles from Trend MD

Plan your remote conference with Sciendo