Loading [MathJax]/extensions/TeX/mathchoice.js
Research article Special Issues

SSMM-DS: A semantic segmentation model for mangroves based on Deeplabv3+ with swin transformer

  • Received: 21 July 2024 Revised: 06 September 2024 Accepted: 23 September 2024 Published: 15 October 2024
  • Mangrove wetlands play a crucial role in maintaining species diversity. However, they face threats from habitat degradation, deforestation, pollution, and climate change. Detecting changes in mangrove wetlands is essential for understanding their ecological implications, but it remains a challenging task. In this study, we propose a semantic segmentation model for mangroves based on Deeplabv3+ with Swin Transformer, abbreviated as SSMM-DS. Using Deeplabv3+ as the basic framework, we first constructed a data concatenation module to improve the contrast between mangroves and other vegetation or water. We then employed Swin Transformer as the backbone network, enhancing the capability of global information learning and detail feature extraction. Finally, we optimized the loss function by combining cross-entropy loss and dice loss, addressing the issue of sampling imbalance caused by the small areas of mangroves. Using GF-1 and GF-6 images, taking mean precision (mPrecision), mean intersection over union (mIoU), floating-point operations (FLOPs), and the number of parameters (Params) as evaluation metrics, we evaluate SSMM-DS against state-of-the-art models, including FCN, PSPNet, OCRNet, uPerNet, and SegFormer. The results demonstrate SSMM-DS's superiority in terms of mIoU, mPrecision, and parameter efficiency. SSMM-DS achieves a higher mIoU (95.11%) and mPrecision (97.79%) while using fewer parameters (17.48M) compared to others. Although its FLOPs are slightly higher than SegFormer's (15.11G vs. 9.9G), SSMM-DS offers a balance between performance and efficiency. Experimental results highlight SSMM-DS's effectiveness in extracting mangrove features, making it a valuable tool for monitoring and managing these critical ecosystems.

    Citation: Zhenhua Wang, Jinlong Yang, Chuansheng Dong, Xi Zhang, Congqin Yi, Jiuhu Sun. SSMM-DS: A semantic segmentation model for mangroves based on Deeplabv3+ with swin transformer[J]. Electronic Research Archive, 2024, 32(10): 5615-5632. doi: 10.3934/era.2024260

    Related Papers:

    [1] Faheem Khan, Tayyaba Arshad, Abdul Ghaffar, Kottakkaran Sooppy Nisar, Devendra Kumar . Numerical solutions of 2D Fredholm integral equation of first kind by discretization technique. AIMS Mathematics, 2020, 5(3): 2295-2306. doi: 10.3934/math.2020152
    [2] Lanyin Sun, Siya Wen . Applications of mixed finite element method based on Bernstein polynomials in numerical solution of Stokes equations. AIMS Mathematics, 2024, 9(12): 35978-36000. doi: 10.3934/math.20241706
    [3] Ruifeng Wu . Bivariate multiquadric quasi-interpolation operators of Lidstone type. AIMS Mathematics, 2023, 8(9): 20914-20932. doi: 10.3934/math.20231065
    [4] Zena Talal Yassin, Waleed Al-Hayani, Ali F. Jameel, Ala Amourah, Nidal Anakira . Solving fuzzy system of Fredholm integro-differential equations of the second kind by using homotopy analysis method. AIMS Mathematics, 2025, 10(1): 1704-1740. doi: 10.3934/math.2025078
    [5] Mustafa Kara . Approximation properties of the new type generalized Bernstein-Kantorovich operators. AIMS Mathematics, 2022, 7(3): 3826-3844. doi: 10.3934/math.2022212
    [6] Liliana Guran, Khushdil Ahmad, Khurram Shabbir, Monica-Felicia Bota . Computational comparative analysis of fixed point approximations of generalized α-nonexpansive mappings in hyperbolic spaces. AIMS Mathematics, 2023, 8(2): 2489-2507. doi: 10.3934/math.2023129
    [7] Ahmed Ayad Khudhair, Saeed Sohrabi, Hamid Ranjbar . Numerical solution of nonlinear complex integral equations using quasi- wavelets. AIMS Mathematics, 2024, 9(12): 34387-34405. doi: 10.3934/math.20241638
    [8] Guorong Zhou, Qing-Bo Cai . Bivariate λ-Bernstein operators on triangular domain. AIMS Mathematics, 2024, 9(6): 14405-14424. doi: 10.3934/math.2024700
    [9] Sihem Semmar, Omar Fetitah, Mohammed Kadi Attouch, Salah Khardani, Ibrahim M. Almanjahie . A Bernstein polynomial approach of the robust regression. AIMS Mathematics, 2024, 9(11): 32409-32441. doi: 10.3934/math.20241554
    [10] Ahmad Alalyani, M. A. Abdou, M. Basseem . The orthogonal polynomials method using Gegenbauer polynomials to solve mixed integral equations with a Carleman kernel. AIMS Mathematics, 2024, 9(7): 19240-19260. doi: 10.3934/math.2024937
  • Mangrove wetlands play a crucial role in maintaining species diversity. However, they face threats from habitat degradation, deforestation, pollution, and climate change. Detecting changes in mangrove wetlands is essential for understanding their ecological implications, but it remains a challenging task. In this study, we propose a semantic segmentation model for mangroves based on Deeplabv3+ with Swin Transformer, abbreviated as SSMM-DS. Using Deeplabv3+ as the basic framework, we first constructed a data concatenation module to improve the contrast between mangroves and other vegetation or water. We then employed Swin Transformer as the backbone network, enhancing the capability of global information learning and detail feature extraction. Finally, we optimized the loss function by combining cross-entropy loss and dice loss, addressing the issue of sampling imbalance caused by the small areas of mangroves. Using GF-1 and GF-6 images, taking mean precision (mPrecision), mean intersection over union (mIoU), floating-point operations (FLOPs), and the number of parameters (Params) as evaluation metrics, we evaluate SSMM-DS against state-of-the-art models, including FCN, PSPNet, OCRNet, uPerNet, and SegFormer. The results demonstrate SSMM-DS's superiority in terms of mIoU, mPrecision, and parameter efficiency. SSMM-DS achieves a higher mIoU (95.11%) and mPrecision (97.79%) while using fewer parameters (17.48M) compared to others. Although its FLOPs are slightly higher than SegFormer's (15.11G vs. 9.9G), SSMM-DS offers a balance between performance and efficiency. Experimental results highlight SSMM-DS's effectiveness in extracting mangrove features, making it a valuable tool for monitoring and managing these critical ecosystems.



    Bernstein polynomials have been used recently to solve some linear as well as nonlinear fuzzy integral equations. These polynomials are positive and their sum is unity. The study of fuzzy integral equations was initially considered by Kaleva [32] to convert the initial value problem for first-order fuzzy differential equations into the equivalent fuzzy Volterra integral equation. Some authors investigated the existence of a unique solution for fuzzy integral equations using the Banach fixed point theorem (see [9,10,28,36,37]). Numerical methods for approximating the solution of fuzzy integral equations have been studied by many authors based on quadrature rules and Pcard's approximations (see [11,12,15,16,23,38,43,48,49,53]). Several numerical iterative approaches have been proposed based on the Picard approximations and other techniques such as the Lagrange interpolation [26], divided and finite differences [35], hybrid block-pulse functions and Taylor series [7], triangular basis functions [8] and block-pulse basis functions [50,53]. Some other existing numerical and analytical methods in literature are: Adomian decomposition [1], Nyström methods [2], Bernstein polynomials [22], fuzzy Haar wavelets [50] and block pulse functions [39], homotopy analysis [34] and homotopy perturbation [5]. Yang and Gong [47] introduced the concept of ill-posedness for the fuzzy Fredholm integral equation of the first kind using the Zadehs decomposition theorem of fuzzy number and obtained the approximate solution for this class of integral equations using regularization method based on classic Tikhonov regularization scheme. Early, Ziari et al. [55] presented an iterative method based on fuzzy Bernstein polynomials for solving nonlinear fuzzy Volterra integral equations. Recently, Ziari et al. [56] proposed an iterative method to solve fuzzy integral equations using generalized quadrature rule. The study of fuzzy integral equations in two dimensions based on iterative technique was started by Sadatrasoul and Ezzati (see [41]). Indeed, they developed the iterative method in [23] for two dimensional fuzzy integral equations. Also, Ezzati and Ziari in [24] proposed a non-iterative numerical method for solving fuzzy Fredholm integral equations in two dimensions based on Bernstein polynomials. Sadatrasoul and Ezzati in [42] presented an iterative method to find approximate solution of linear and nonlinear two dimensional Hammerstein fuzzy integral equations by considering optimal quadrature formula for fuzzy-number-valued functions of Lipschitz type. In [16], Bica and Popescu constructed the fuzzy trapezoidal cubature rule for fuzzy functions of Lipschitz type to approximate the solution of nonlinear fuzzy Fredholm integral equations in two variables. Recently, Akhavan et al. [3] developed an iterative approach and combining trapezoidal and midpoint formulas for such equations. In [33], the existence and uniqueness results for fuzzy Fredholm-Volterra integral have been investigated and the authors proposed an iterative numerical approach for these type of equations based on bivariate triangular functions. In this paper, we approximate the integral of two dimensional fuzzy function by the bivariate fuzzy Bernstein polynomials and investigate the error analysis of the numerical method. Also, an iterative procedure is constructed based on two dimensional fuzzy Bernstein polynomials for solving two dimensional nonlinear fuzzy Fredholm integral equations,

    F(x,y)=f(x,y)dcbaH(x,y,s,t)G(F(s,t))dsdt,(s,t)[a,b]×[c,d]. (1.1)

    The main advantage of proposed method is the reduction of computational cost with respect to existing methods in the literature such as iterative methods based on fuzzy trapezoidal cubature rule in [16] and [41]. For instance, the number of function evaluations of the proposed method at the knots is n2 for m=n and the same factor for methods in [16] and [41] equals to 4n2 for m=n, thus, the proposed approach is reasonable from aspect of reduction of computational cost in comparison with the existing methods. We provide some numerical experiments to confirm the numerical results with theoretical results. Finally, some concluding remarks will be presented.

    Definition 2.1. (See [20].) A fuzzy number is a function w:R[0,1] having the properties:

    (1) w is normal, that is x0R such that w(x0)=1,

    (2) w is fuzzy convex set

    (i.e.w(λx+(1λ)y)min{w(x),w(y)}x,yR,λ[0,1]),

    (3) w is upper semicontinuous on R,

    (4) the ¯{xR:w(x)>0} is compact set.

    The set of all fuzzy numbers is denoted by E.

    Definition 2.2. (See [27].) An arbitrary fuzzy number is represented, in parametric form, by an ordered pair of functions (w_(r),¯w(r)),0r1, which satisfy the following requirements:

    (1) w_(r) is a bounded left continuous nondecreasing function over [0, 1],

    (2) ¯w(r) is a bounded left continuous nonincreasing function over [0, 1],

    (3) w_(r)¯w(r),0r1.

    For addition and scalar multiplication of fuzzy numbers in fuzzy set space E we have:

    (1) w1w2=(w1_(r)+w2_(r),¯w1(r)+¯w2(r)),

    (2) (λw1)={(λw1_(r),λ¯w1(r))λ0,(λ¯w1(r),λw1_(r))λ<0.

    Definition 2.3. (See [31].) For given fuzzy numbers w1=(w1_(r),¯w1(r)) and w2=(w2_(r),¯w2(r)), the quantity

    D(w1,w2)=supr[0,1]max{|w1_(r)w2_(r)|,|¯w1(r)¯w2(r)|}

    is the Hausdorff distance between fuzzy numbers w1 and w2.

    Lemma 2.1. (See [46].) If w1,w2,w3,w4E and kR, then we have:

    (1)D(w1w3,w2w3)=D(w1,w2),

    (2)D(kw1,kw2)=|k|D(w1,w2),

    (3)D(w1w2,w3w4)D(w1,w3)+D(w2,w4),

    (4)D(w1w2,˜0)D(w1,˜0)+D(w2,˜0).

    In [46], it is proved that (E,D) is a complete metric space.

    Lemma 2.2. (see [4].) For any k1,k2R with k1k20 and any wE we have D(k1w,k2w)=|k1k2|D(w,˜0).

    Remark 2.1. The property (4) in Lemma (2.1) provide the definition of a function .:ER+ by w1=D(w1,˜0), which has the properties of the usual norms. In [5] the properties of this function are exhibited as follows:

    (i) w10,w1E, and w1=0 iff w1=˜0,

    (ii) λ.w1=|λ|w1 and w1w2w1+w2,w1,w2E,λR,

    (iii) |w1w2|D(w1,w2) and D(w1,w2)w1+w2w1,w2E.

    Definition 2.4. (See [25].) A fuzzy-real-number-valued function f:[a,b]E is said to be continuous in t0[a,b], if for given ε>0 there exists a δ>0 such that |xx0|<δ. implies that D(f(x),f(x0))<ε, for all x[a,b] and We say that f is fuzzy continuous on [a,b] if f is continuous at each t0[a,b], and we will use the notation C([a,b],E) for the space of all continuous fuzzy functions.

    Definition 2.5. (See [29].) Let f:[a,b]E be given. The function f is fuzzy-Riemann integrable to I(f)E if for any ε>0, there exists δ>0 such that for any division P={[u,v];ξ} of [a,b] with the norms Δ(p)<δ, we have

    D(P(vu)f(ξ),I(f))<ε,

    where denotes the fuzzy summation. In this case it is denoted by I(f)=(FR)baf(x)dx.

    Lemma 2.3. (See [29].) If f,g:[a,b]RE are fuzzy continuous functions, then the function F:[a,b]R+ by F(x)=D(f(x),g(x)) is continuous on [a,b] and

    D((FR)baf(x)dx,(FR)bag(x)dx)baD(f(x),g(x))dx.

    Theorem 2.1. (See [30].) If f,g:[a,b]E are (FR) integrable fuzzy functions, and α, β are real numbers, then

    (FR)ba(αf(x)βg(x))dx=α(FR)baf(x)dxβ(FR)bag(x)dx.

    Definition 2.6. (See [11].) For L0, a function f:[a,b]E is L-Lipschitz if

    D(f(x),f(y))L|xy|

    for any x,y[a,b]. A function F:EE is Lipschitz if there exists L0 such that D(F(u),F(v))LD(u,v), for any u,vE.

    According to [11], any Lipschitz function is continuous.

    For fCF[a,b], let us consider the Bernstein-type fuzzy polynomials

    B(F)n(f)(x)=nk=0f(a+k(ba)n)pn,k(x),nN,x[a,b],

    where Pn,k(x)(k=0,...,n) are Bernstein polynomials of degree n defined on [a,b] as:

    Pn,k(x)=(nk)(xa)k(bx)nx(ba)n,k=0,...,n.

    It is obvious that Pn,k(x)0,x[a,b] and Pn,0(x),Pn,1(x),...,Pn,n(x) are linearly independent algebraic polynomials of degree n and nk=0pn,k(x)=1.

    The following definitions are related to fuzzy-number-valued functions in two variables.

    Definition 2.9.(See [3].) Let f:[a,b]×[c,d]E be given. The function f is fuzzy-Riemann double integrable to I(f)E if for any ε>0, there exists δ>0 such that for any partition Δ=(Δx,Δy) of the region [a,b]×[c,d], with Δx: a=x0<x1<xm1<xm=b and Δy:c=y0<y1<<yn1<yn=d and any set of intermediate points {(ξi,ηj):i=1,2,...,m,j=1,2,...,n}, with norm υ(Δ)=max{|xixi1|+|yjyj1|:i=1,,m,j=1,,n}<δ, we have

    D(mi=1nj=1(xixi1)(yjyj1)f(ξi,ηj),I(f))<ε, (2.1)

    In this case, the fuzzy number I(f) will be denoted by

    I(f)=(FR)dc(FR)(baf(s,t)ds)dt. (2.2)

    Definition 2.10. (See [40].) A function f:[a,b]×[c,d]E is called:

    (i) continuous in (x0,y0)[a,b]×[c,d] if for any ε>0 there exists δ>0 such that for any (x,y)[a,b]×[c,d] with |xx0|<δ, |yy0|<δ we have

    D(f(x,y),f(x0,y0))<ε.

    The function f is continuous on [a,b]×[c,d] if it is continuous in each (x,y)[a,b]×[c,d].

    (ii) bounded if there exists M0 such that

    D(f(x,y),˜0)M,(x,y)[a,b]×[c,d].

    The set of all continuous functions f:[a,b]×[c,d]E is denoted by C([a,b]×[c,d],E).

    Lemma 2.4. (See [40].) If fC([a,b]×[c,d],E) then

    (FR)dc((FR)baf(x,y)dx)dy

    exists and

    ((FR)dc((FR)baf(x,y)dx)dy)_(r)=dc(baf(x,y,r)_dx)dy
    ¯((FR)dc((FR)baf(x,y)dx)dy)(r)=dc(ba¯f(x,y,r)dx)dy.

    Lemma 2.5. If f,g:[a,b]×[c,d]E are continuous fuzzy functions then the function φ:[a,b]×[c,d]R+ defined by φ(s,t)=D(f(x,y),g(x,y)) is continuous on [a,b]×[c,d] and

    D((FR)dc((FR)baf(x,y)dx)dy,(FR)dc((FR)bag(x,y)dx)dy)
    dc(baD(f(x,y),g(x,y))dx)dy.

    Proof. For simplicity we consider only m=n, where the m and n are the number of points in the partition of [a,b] and [c,d], respectively, the case of mn follows similarly. Firstly, we prove that the function φ(s,t)=D(f(s,t),g(s,t)) be continuous in every point (s0,t0)[a,b]×[c,d], for this purpose, we let {(sn,tn)}n1,(sn,tn)[a,b]×[c,d], such that limn(sn,tn)=(s0,t0). In this case, we have:

    D(f(sn,tn),g(sn,tn))D(f(sn,tn),f(s0,t0))+D(f(s0,t0),g(s0,t0))+D(g(s0,t0),g(sn,tn)),

    and on the other hand, we have:

    D(f(s0,t0),g(s0,t0))D(f(s0,t0),f(sn,tn))+D(f(sn,tn),g(sn,tn))+D(g(sn,tn),g(s0,t0)).

    Taking to the limit when n, and according to the continuity of f and g we obtain:

    limnD(f(sn,tn),g(sn,tn))=D(f(s0,t0),g(s0,t0)),

    that is φ is continuous at each (s0,t0)[a,b]×[c,d]. Now, let Pn={([sn1,sn];ξn)},nN and Qn={([tn1,tn];ηn)},nN be two sequences of partitions of [a,b] and [c,d] with Δ(Pn)0, when n and Δ(Qn)0, when n, respectively.

    So, we have:

    D((FR)dc(FR)(baf(s,t)ds)dt,(FR)dc(FR)(bag(s,t)ds)dt)
    D((FR)dc(FR)(baf(s,t)ds)dt,ni=1nj=1(xixi1)(yjyj1)f(ξi,ηj))+
    D(ni=1nj=1(xixi1)(yjyj1)f(ξi,ηj),ni=1nj=1(xixi1)(yjyj1)g(ξi,ηj))+
    D(ni=1nj=1(xixi1)(yjyj1)g(ξi,ηj),(FR)dc(FR)(bag(s,t)ds)dt)
    D((FR)dc(FR)(baf(s,t)ds)dt),,ni=1nj=1(xixi1)(yjyj1)f(ξi,ηj))+
    ni=1nj=1D(f(ξi,ηj),g(ξi,ηj))(xixi1)(yjyj1)+D(ni=1nj=1(xixi1)(yjyj1)g(ξi,ηj),
    (FR)dc(FR)(bag(s,t)ds)dt)

    Taking to the limit when n, we get:

    D((FR)dc(FR)(baf(s,t)ds)dt,(FR)dc(FR)(bag(s,t)ds)dt)
    dcbaD(f(s,t),g(s,t))dsdt.

    Thus, the proof is complete.

    Moreover, it can be proved that any continuous function f:[a,b]×[c,d]E is bounded.

    Consider a fuzzy-number-valued function f:[a,b]×[c,d]E having the following Lipschitz property: there exist L1,L20 such that

    D(f(x1,y1),f(x2,y2))L1|x1x2|+L2|y1y2| (2.3)

    for all x1,x2[a,b], y1,y2[c,d].

    Similar to the one dimensional case, for fCF([a,b]×[c,d]) we express the fuzzy two- dimensional Bernstein operators as follows:

    Definition 2.10. (See [4].) Two dimensional fuzzy Bernstein operator on the region [a,b]×[c,d] is defined as follows:

    B(F)m,n(f)(x,y)=mi=0nj=0f(xi,yj)Pm,i(x)Pn,j(y),(x,y)[a,b]×[c,d],(m,n)N2, (2.4)

    where xi=a+i(ba)m,(i=0,...,m),yj=c+j(dc)n,(j=0,...n) and Pm,i(x),(i=0,...,m),Pn,j(x),(j=0,...,n) are Bernstein polynomials of degrees m and n respectively, moreover the polynomials Pm,i(x) and Pn,ji,j, are defined on intervals [a,b] and [c,d], respectively.

    Let φi,j(x,y)=Pm,i(x).Pn,j(y), it is obvious that ϕi,j(x,y)0,(x,y)[a,b]×[c,d],i,j, and φ0,0(x,y),φ0,1(x,y),...,φm,n(x,y) are linearly independent, and

    mi=0nj=0φi,j(s,t)=1. (2.5)

    The fuzzy function fCF([a,b]×[c,d]) can be approximated using fuzzy Two dimensional Bernstein-type polynomials as

    f(s,t)mi=0nj=0f(si,tj)Pm,i(s)Pn,j(t),

    where si=a+i(ba)m,(i=0,...,m),tj=c+j(dc)n,(j=0,...n). Hence, the approximate value of the integral of fuzzy function can be obtained as follows:

    (FR)dc(FR)baf(s,t)dsdt(FR)dc(FR)bami=0nj=0f(si,tj)Pm,i(s)Pn,j(t)dsdt=mi=0nj=0f(si,tj)dcbaPm,i(s)Pn,j(t)dsdt.

    In [21], the derivative of Bernstein polynomials on interval [a,b] for fixed m is expressed as follows:

    Pm,i(s)=mba(Pm1,i1(s)+Pm1,i(s)).

    By integrating from the above equality, we obtain:

    baPm1,i1(s)ds=baPm1,i(s)ds,

    namely, for fixed m all Bernstein basis functions have the same definite integral over the interval [a, b]. Since mi=0pm,i(s)=1 we have:

    baPm,i(s)ds=bam+1i=0,1,...,m.

    Similarity for two dimensional Bernstein polynomials, we obtain:

    dcbaPm,i(s)Pn,j(t)dsdt=(ba)(dc)(m+1)(n+1),i,j. (3.1)

    So, we have the following approximation for double integral of two dimensional fuzzy function:

    (FR)dc(FR)baf(s,t)dsdt(ba)(dc)(m+1)(n+1)mi=0nj=0f(si,tj), (3.2)

    where si=a+i(ba)m,(i=0,...,m),tj=c+j(dc)n,(j=0,...n).

    In the following theorem, we obtain the error estimate of the above approximation.

    Theorem 3.1. Let fCF([a,b]×[c,d]) be a Lipschitzian function. Then we have:

    Σ=D((FR)dc(FR)baf(s,t)dsdt,(ba)(dc)(m+1)(n+1)mi=0nj=0f(si,tj))(L1(ba)2(dc)2m+L2(ba)(dc)22n).

    Proof. According to (2.5), we have:

    Σ=D((FR)dc(FR)baf(s,t)dsdt,(ba)(dc)(m+1)(n+1)mi=0nj=0f(si,tj))=D((FR)dc(FR)bami=0nj=0φi,j(s,t)f(s,t)dsdt,(FR)dc(FR)bami=0nj=0φi,j(s,t)f(si,tj)dsdt).

    where φi,j(s,t)=Pm,i(s).Pn,j(t). Then, applying the parts of 2 and 3 of Lemma 2.1, we obtain:

    Σdcbami=0nj=0φi,j(s,t)D(f(s,t),f(si,tj))dsdt.

    As regards, f satisfies in Lipschitz condition namely, inequality (2.3) holds for f, we obtain:

    Σdcbami=0nj=0φi,j(s,t)(L1|ssi|+L2|ttj|)dsdt.

    It is well known following inequality which is expressed in [29]:

    mi=0|ssi|Pm,i(s)ba2m,s[a,b],mN. (3.3)

    According to φi,j(s,t)=Pm,i(s).Pn,j(t) and considering the mi=0Pm,i(s)=1,nj=0Pn,j(t)=1, and inequality (3.3), we obtain:

    Σ=D((FR)dc(FR)baf(s,t)dsdt,(ba)(dc)(m+1)(n+1)mi=0nj=0f(si,tj))(L1(ba)2(dc)2m+L2(ba)(dc)22n).

    Here, we consider the two dimensional nonlinear fuzzy Fredholm integral Eq. (1.1), where H(x, y, s, t) is a crisp kernel function over ([a,b]×[c,d])2, f,F are continuous fuzzy-number-valued functions and G:EE is a continuous fuzzy function. We assume that H is continuous and therefore it is uniformly continuous with respect to (s,t) and there exists MH>0, such that MH=maxax,sb,cy,td|H(x,y,s,t)|.

    Let Ω={f:[a,b]×[c,d]E;f is continuous} be the space of the two-dimensional fuzzy continuous functions with the metric D(f,g)=supasb,ctdD(f(s,t),g(s,t)), for f,gΩ. In the following theorem, sufficient conditions for the existence of an unique solution of Eq (1.1) are given.

    Theorem 4.1. (See [41].) Let H(x,y,s,t) be continuous and positive for ax,sb,cy,td and f:[a,b]×[c,d]E be continuous on [a,b]×[c,d]. Moreover assume that there exists L>0, such that

    D(G(F1(s,t)),G(F2(s,t)))L.D(F1(s,t),F2(s,t)),(s,t)[a,b]×[c,d],F1,F2Ω.

    If B=MHL(ba)(dc)<1, then the iterative procedure for kN

    F0(x,y)=f(x,y),Fk(x,y)=f(x,y)(FR)dc(FR)baH(x,y,s,t)G(Fk1(s,t))dsdt,(x,y)[a,b]×[c,d]. (4.1)

    converges to the solution of F of (1.1). In addition, the following error bound holds:

    D(F,Fk)Bk1BD(F1,F0),kN. (4.2)

    where Ω is the set of fuzzy continuous functions on I=[a,b]×[c,d].

    Remark 4.1. If F0=f the error estimate (4.2), becomes:

    D(F,Fk)Bk+1L(1B)(Lfϝ+M0),kN, (4.3)

    where M0=sup(x,y)IG(˜0)F.

    Proof. It is observed that

    D(F0,F1)=sup(x,y)ID(f(x,y),f(x,y)(FR)dc(FR)baH(x,y,s,t)G(F0(s,t))dsdt)sup(x,y)IdcbaD(˜0,H(x,y,s,t)G(f(s,t)))dsdtdcbasup(x,y)Isup(s,t)I|H(x,y,s,t)|D(˜0,G(f(s,t)))dsdtMHdcbasup(s,t)I[D(˜0,G(˜0))+D(G(˜0),G(f(s,t)))]dsdtMH(ba)(dc)[sup(s,t)IG(˜0)ϝ+LD(˜0,f)]BL(M0+Lfϝ),

    and hence, it obtains the inequality (4.3).

    Now, we introduce the numerical method to find the approximate solution of the two dimensional nonlinear fuzzy Fredholm integral equation (1.1). In this way, we consider the following uniform partitions of the region :

    (4.4)

    with . Then according to (3.2), the following iterative procedure gives the approximate solution of Eq (1.1) in the point using two dimensional fuzzy Bernstein-type polynomials:

    (4.5)

    In this section, we investigate the convergence of the iterative proposed method to the solution of Eq (1.1) under the following conditions:

    (i) is fuzzy continuous;

    (ii) is continuous;

    (iii) There exist such that

    for any

    (iv) There exists such that

    where ;

    (v) where is as given in the above item and is such that ;

    (vi) There exist such that

    for any

    (vii) There exist such that

    for any

    Firstly, we prove that functions for all are Lipschitzian.

    Lemma 5.1. Consider the iterative procedure (4.5). Under the conditions (i)–(vii) then functions for all are Lipschitzian.

    Proof. Using the properties of the distance between fuzzy numbers stated in Lemma 2.1 we obtain

    Consequently, using conditions (iii) and (vii), we obtain:

    (5.1)

    By applying the properties of the norm function in Remark 1, we obtain

    (5.2)

    By using again the properties of the norm function in Remark 1, we have

    By applying the inequality (5.2) into the above inequality, we have:

    Taking supremum from the above inequality for , it follows that

    By successive substitutions on the above inequality, we obtain

    (5.3)

    Since, for all , we obtain:

    (5.4)

    Taking supremum from the inequality (5.2), it follows that

    Now, by taking into account the inequality (5.4), from the above inequality we get:

    (5.5)

    Finally, by considering the above inequality, from (5.1) we obtain:

    By supposing and , we have:

    Thus, the functions for all are Lipschitzian.

    Now, we prove an interesting result about the satisfying functions in Lipschitz condition which is used in the proof of the main result.

    Lemma 5.2. Consider the iterative procedure (4.5). Under the conditions (i)–(vii) and supposing functions are Lipschitzian then the functions are Lipschitzian.

    Proof. Using fuzzy Distance, we have:

    Using part (2) of Lemma 2.1, Lemma 2.2 and condition (iii), we obtain:

    According to Lemma 5.1 and using condition (vi), we get:

    (5.6)

    By applying the inequality (5.5) from the proof of Lemma 5.1, we obtain

    Now, by taking into account the above inequality, from (5.6) we get:

    By assuming and we have:

    Hence, the functions for all are Lipschitzian.

    Theorem 5.1. Under the conditions (i)–(vii) the iterative procedure Eq (4.5) converges to the unique solution of Eq (1.1), F, and its error estimate is as follows:

    where .

    Proof. Let and for all . Since

    we have

    Regarding to part (2) of Lemma 2.1, Theorem 3.1 and condition (iii), we have:

    Applying Lemma 5.1 and Theorem 3.1 for second part of above expression, we obtain:

    Taking supremum from the above inequality for , and according to , we deduce:

    Now, since

    we conclude that

    Using again part (2) of Lemma 2.1, Theorem 3.1, condition (iii) and Lemma 5.1, we obtain:

    Then, we have:

    By induction, for , we obtain:

    Therefore, we have:

    Since , we conclude that

    (5.7)

    Using Eqs (4.3) and (5.7), we obtain:

    Remark 5.1. Since , it is easy to show that

    this result shows that the proposed method is convergent.

    For the iterative numerical method, it is more suitable investigating the stability of the obtained numerical solution with respect to the choice of the first iteration. So, in order to study the numerical stability of the iterative method (4.5) with respect to small changes in the starting approximation, we consider an another initial iteration term such that there exists for which , for all So, we have the new sequence of successive approximations as:

    (6.1)

    Using the same iterative method, the terms of produced sequence are:

    (6.2)

    Definition 6.1. We say that the method of successive approximations applied to solve Eq. (1.1) is numerically stable with respect to the choice of the first iteration iff there exist positive numbers and constants which are independent by step-sizes and respectively, such that

    (6.3)

    Theorem 6.1. Under the conditions of Theorem 5.1, the iterative method (6.2) is numerically stable with respect to the choice of the first iteration.

    Proof. Clearly, we observe that:

    (6.4)

    Using Eq (5.7), we obtain:

    (6.5)

    Also, by similar reasoning we have:

    (6.6)

    In order to obtain the upper bound of , we observe that

    and thus

    Therefor, we obtain:

    By induction for , we get:

    According to , this inequality becomes:

    Then, from eq-stab1, we get:

    (6.7)

    By comparing inequalities (6.3) and (6.7), we deduce that

    So, the proof is complete.

    In this section, we present numerical results obtained by applying the proposed algorithm on two examples in order to demonstrate the validity and accuracy of the method. The approximate solution is calculated for different values of , and .

    Example 1. Consider

    (7.1)

    where

    The above fuzzy equation has the exact solution

    The results of the proposed iterative algorithm in given points are shown in terms of errors in Tables 1 and 2.

    Table 1.  The accuracy on the level sets for Example 1 in .
    k=10, m=10, n=10 k=15, m=30, n=30
    r-level
    0 0.0000E+0 7.7426E-3 0.0000E+0 2.4452E-3
    0 0.0000E+0 7.7426E-3 0.0000E+0 2.4452E-3
    0.25 1.0452E-4 5.7966E-3 3.3656E-5 1.8366E-3
    0.50 4.2630E-4 4.1667E-3 1.3695E-4 1.3242E-3
    0.75 9.7850E-4 2.8324E-3 3.1356E-4 9.0278E-4
    1 1.7754E-3 1.7754E-3 5.6742E-4 5.6742E-4

     | Show Table
    DownLoad: CSV
    Table 2.  The accuracy on the level sets for Example 1 in .
    k=10, m=10, n=10 k=15, m=30, n=30
    r-level
    0 0.0000E+0 8.0781E-3 0.0000E+0 4.7666E-3
    0.25 2.3896E-4 6.9781E-3 8.5789E-5 3.8241E-3
    0.50 8.6241E-4 5.4650E-3 3.4468E-4 2.9150E-3
    0.75 1.7319E-3 3.6661E-3 7.7281E-4 2.0815E-3
    1 2.7111E-3 2.7111E-3 1.3583E-3 1.3583E-3

     | Show Table
    DownLoad: CSV

    This method well performs for linear two dimensional fuzzy integral equation.

    Example 2. Consider

    (7.2)

    where

    The above fuzzy equation has the exact solution

    The results of the proposed iterative algorithm in given points are shown in terms of errors in Tables 3 and 4.

    Table 3.  The accuracy on the level sets for Example 2 in .
    k=10, m=10, n=10 k=10, m=30, n=30
    r-level
    0 0.0000E+0 1.7731E-3 0.0000E+0 5.9032E-4
    0.25 2.2163E-4 1.5514E-3 7.3790E-5 5.1653E-4
    0.50 4.4326E-4 1.3298E-3 1.4758E-4 4.4274E-4
    0.75 6.6489E-4 1.1082E-3 2.2137E-4 3.6895E-4
    1 8.8652E-4 8.8652E-4 2.9516E-4 2.9516E-4

     | Show Table
    DownLoad: CSV
    Table 4.  The accuracy on the level sets for Example 2 in .
    k=10, m=10, n=10 k=10, m=30, n=30
    r-level
    0 0.0000E+0 2.6347E-3 0.0000E+0 7.5628E-4
    0.25 3.2933E-4 2.3053E-3 9.4535E-5 6.6174E-4
    0.50 6.5867E-4 1.9760E-3 1.8907E-4 5.6721E-4
    0.75 9.8800E-4 1.6467E-3 2.8361E-4 4.7268E-4
    1 1.3173E-3 1.3173E-3 3.7814E-4 3.7814E-4

     | Show Table
    DownLoad: CSV

    In this paper, we have developed an numerical iterative method for the solution of two dimensional fuzzy Fredholm integral equations of the second kind. The method presented in this work was applied Picard iterations and numerical approximation to evaluate the iterated integral based on bivariate Bernstein polynomials. Moreover, we have considered the conditions to prove the sequence of approximations of the solution at given point converge to the exact solution. The concept of numerical stability is defined based on the choice of the first iteration and then the numerical stability of the proposed iterative algorithm is proven. Finally, some numerical examples are included to examine the accuracy and the convergence of the proposed method.

    The authors declare that there is no conflict of interest.



    [1] N. C. Duke, J. Meynecke, S. Dittmann, A. M. Ellison, K. Anger, U. Berger, et al., A world without mangroves?, Science, 317 (2007), 41–42. https://doi.org/10.1126/science.317.5834.41b doi: 10.1126/science.317.5834.41b
    [2] P. L. Biswas, S. R. Biswas, Mangrove forests: Ecology, management, and threats, Life Land, (2019), 1–14. https://doi.org/10.1007/978-3-319-71065-5_26-1 doi: 10.1007/978-3-319-71065-5_26-1
    [3] C. Giri, Frontiers in global mangrove forest monitoring, Remote Sens., 15 (2023), 3852. https://doi.org/10.3390/rs15153852 doi: 10.3390/rs15153852
    [4] S. C. Basha, An overview on global mangroves distribution, Indian J. Geo Mar. Sci., 47 (2018), 766–772.
    [5] Y. Sun, D. Zhao, W. Guo, Y. Gao, X. Su, B. Wei, A review on the application of remote sensing in mangrove ecosystem monitoring, Acta Ecol. Sin., 33 (2013), 4523–4538. https://doi.org/10.5846/stxb201205150715 doi: 10.5846/stxb201205150715
    [6] R. Vidhya, D. Vijayasekaran, M. Ahamed Farook, S. Jai, M. Rohini, A. Sinduja, Improved Classification of mangroves health status using hyperspectral remote sensing data, Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci., 8 (2014), 667–670. https://doi.org/10.5194/isprsarchives-XL-8-667-2014 doi: 10.5194/isprsarchives-XL-8-667-2014
    [7] T. D. Pham, D. T. Bui, K. Yoshino, N. N. Le, Optimized rule-based logistic model tree algorithm for mapping mangrove species using ALOS PALSAR imagery and GIS in the tropical region, Environ. Earth Sci., 77 (2018), 159. https://doi.org/10.1007/s12665-018-7373-y doi: 10.1007/s12665-018-7373-y
    [8] L. Li, W. Liu, Y. Tao, X. Xu, W. Fu, J. Dong, Diffusion dynamics and driving forces of Spartina alterniflora in the Guangxi Shankou Mangrove Reserve, Acta Ecol. Sin., 41 (2021), 6814–6824.
    [9] J. Cao, K. Liu, L. Zhuo, L. Liu, Y. Zhu, L. Peng, Combining UAV-based hyperspectral and LiDAR data for mangrove species classification using the rotation forest algorithm, Int. J. Appl. Earth Obs. Geoinformation, 102 (2021), 102414. https://doi.org/10.1016/j.jag.2021.102414 doi: 10.1016/j.jag.2021.102414
    [10] M. Di Cicco, C. Potena, G. Grisetti, A. Pretto, Automatic model based dataset generation for fast and accurate crop and weeds detection, in 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), (2017), 5188–5195. https://doi.org/10.1109/IROS.2017.8206408
    [11] D. Lee, C. Kim, S. Kim, M. Cho, W. Han, Autoregressive image generation using residual quantization, in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2022), 11513–11522. https://doi.org/10.1109/CVPR52688.2022.01123
    [12] L. Wang, R. Li, C. Zhang, S. Fang, C. Duan, X. Meng, et al., UNetFormer: A UNet-like transformer for efficient semantic segmentation of remote sensing urban scene imagery, ISPRS J. Photogramm. Remote Sens., 190 (2022), 196–214. https://doi.org/10.1016/j.isprsjprs.2022.06.008 doi: 10.1016/j.isprsjprs.2022.06.008
    [13] X. Yuan, J. Shi, L. Gu, A review of deep learning methods for semantic segmentation of remote sensing imagery, Expert Syst. Appl., 169 (2021), 114417. https://doi.org/10.1016/j.eswa.2020.114417 doi: 10.1016/j.eswa.2020.114417
    [14] A. A. Aleissaee, A. Kumar, R. M. Anwer, S. Khan, H. Cholakkal, G. Xia, et al., Transformers in Remote Sensing: A Survey, Remote Sens., 15 (2023), 1860. https://doi.org/10.3390/rs15071860 doi: 10.3390/rs15071860
    [15] L. Wan, H. Zhang, G. Lin, H. Lin, A small-patched convolutional neural network for mangrove mapping at species level using high-resolution remote-sensing image, Ann. GIS, 25 (2019), 45–55. https://doi.org/10.1080/19475683.2018.1564791 doi: 10.1080/19475683.2018.1564791
    [16] G. M. D. S. Moreno, O. A. D. Carvalho Júnior, O. L. F. D. Carvalho, T. C. Andrade, Deep semantic segmentation of mangroves in Brazil combining spatial, temporal, and polarization data from Sentinel-1 time series, Ocean Coastal Manage., 231 (2023), 106381. https://doi.org/10.1016/j.ocecoaman.2022.106381 doi: 10.1016/j.ocecoaman.2022.106381
    [17] Y. Fan, Q. Zeng, Z. Mei, W. Hu, Semantic segmentation for mangrove using spectral indices and self-attention mechanism, in 2022 7th International Conference on Signal and Image Processing (ICSIP), (2022), 436–441. https://doi.org/10.1109/ICSIP55141.2022.9886553
    [18] C. Xu, J. Wang, Y. Sang, K. Li, J. Liu, G. Yang, An effective deep learning model for monitoring mangroves: A case study of the Indus Delta, Remote Sens., 15 (2023), 2220. https://doi.org/10.3390/rs15092220 doi: 10.3390/rs15092220
    [19] C. Kuenzer, A. Bluemel, S. Gebhardt, T. V. Quoc, S. Dech, Remote sensing of mangrove ecosystems: A Review, Remote Sens., 3 (2011), 878–928. https://doi.org/10.3390/rs3050878 doi: 10.3390/rs3050878
    [20] K. Maurya, S. Mahajan, N. Chaube, Remote sensing techniques: mapping and monitoring of mangrove ecosystem-A review, Complex Intell. Syst., 7 (2021), 2797–2818. https://doi.org/10.1007/s40747-021-00457-z doi: 10.1007/s40747-021-00457-z
    [21] C. Giri, Observation and monitoring of mangrove forests using remote sensing: Opportunities and challenges, Remote Sens., 8 (2016), 783. https://doi.org/10.3390/rs8090783 doi: 10.3390/rs8090783
    [22] T. V. Tran, R. Reef, X. Zhu, A review of spectral indices for mangrove remote sensing, Remote Sens., 14 (2022), 4868. https://doi.org/10.3390/rs14194868 doi: 10.3390/rs14194868
    [23] C. Liu, P. Sun, S. Liu, A review of plant spectral reflectance response to water physiological changes, Chin. J. Plant Ecol., 40 (2016), 80–91. https://doi.org/10.17521/cjpe.2015.0267 doi: 10.17521/cjpe.2015.0267
    [24] J. Xue, B. Su, Significant remote sensing vegetation indices: A review of developments and applications, J. Sens., (2017). https://doi.org/10.1155/2017/1353691 doi: 10.1155/2017/1353691
    [25] B. Gao, NDWI-A normalized difference water index for remote sensing of vegetation liquid water from space, Remote Sens. Environ., 58 (1996), 257–266. https://doi.org/10.1016/S0034-4257(96)00067-3 doi: 10.1016/S0034-4257(96)00067-3
    [26] K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning for image recognition, in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2016), 770–778. https://doi.org/10.1109/CVPR.2016.90
    [27] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, et al., Attention is all you need, in Proceedings of the 31st International Conference on Neural Information Processing Systems, (2017), 6000–6010.
    [28] X. Li, X. Sun, Y. Meng, J. Liang, F. Wu, J. Li, Dice loss for data-imbalanced NLP tasks, in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, (2020), 465–476. https://doi.org/10.18653/v1/2020.acl-main.45
    [29] L. Chen, Y. Zhu, G. Papandreou, F. Schroff, H. Adam, Encoder-decoder with atrous separable convolution for semantic image segmentation, in Computer Vision-ECCV 2018, 11211 (2018), 833–851. https://doi.org/10.1007/978-3-030-01234-2_49
    [30] C. Shorten, T. M. Khoshgoftaar, A survey on image data augmentation for deep learning, J. Big Data, 6 (2019). https://doi.org/10.1186/s40537-019-0197-0 doi: 10.1186/s40537-019-0197-0
    [31] J. Long, E. Shelhamer, T. Darrell, Fully convolutional networks for semantic segmentation, in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2015), 3431–3440. https://doi.org/10.1109/CVPR.2015.7298965
    [32] H. Zhao, J. Shi, X. Qi, X. Wang, J. Jia, Pyramid scene parsing network, in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (2017), 6230–6239. https://doi.org/10.1109/CVPR.2017.660
    [33] Y. Yuan, X. Chen, J. Wang, Object-contextual representations for semantic segmentation, in Computer vision-ECCV 2020, 12351 (2020), 173–190. https://doi.org/10.1007/978-3-030-58539-6_11
    [34] T. Xiao, Y. Liu, B. Zhou, Y. Jiang, J. Sun, Unified perceptual parsing for scene understanding, in Computer Vision-ECCV 2018: 15th European Conference, (2018), 432–448. https://doi.org/10.1007/978-3-030-01228-1_26
    [35] E. Xie, W. Wang, Z. Yu, A. Anandkumar, J. M. Alvarez, P. Luo, SegFormer: Simple and efficient design for semantic segmentation with transformers, in Proceedings of the 35th International Conference on Neural Information Processing Systems, 34 (2021), 12077–12090.
  • This article has been cited by:

    1. Chandra Shekhar Pant, H. S. Nayal, 2024, Chapter 58, 978-981-97-5503-5, 488, 10.1007/978-981-97-5504-2_58
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(800) PDF downloads(48) Cited by(0)

Figures and Tables

Figures(8)  /  Tables(5)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog