
Typhoons cause significant damage to coastal and inland areas, making the accurate classification of typhoon cloud images essential for effective monitoring and forecasting. While integrating multimodal data from different satellites can improve classification accuracy, existing methods often rely on aligned images and fail to account for radiometric and structural differences, leading to performance degradation during image fusion. In registration methods designed to address this issue, two-stage approaches inaccurately estimate deformation fields, while one-stage methods typically overlook radiometric differences between typhoon cloud images. Additionally, fusion methods suffer from inherent noise accumulation and insufficient cross-modal feature utilization due to cascaded structures. To address these issues, this study proposed a coarse-to-fine registration and fusion network (CoReFuNet) that integrated a one-stage registration module with a cross-modal fusion module for multimodal typhoon cloud image classification. The registration module adopted a one-stage coarse-to-fine strategy, using cross-modal style alignment to address radiometric difference and global spatial registration by affine transformations to resolve positional differences. Bidirectional local feature refinement (BLFR) then ensured precise adjustments, facilitating fine registration by evaluating feature points in each image. Following registration, the fusion module employed a dual-branch alternating enhancement (DAE) approach, which reduced noise by learning cross-modal mapping relationships and applying feedback-based adjustments. Additionally, a cross-modal feature interaction (CMFI) module merged low-level, high-level, intra-modal, and intermodal features through a residual structure, minimizing modality differences and maximizing feature utilization. Experiments on the FY-HMW (Feng Yun-Himawari) dataset, constructed using data from the Feng Yun and Himawari satellites, showed that the CoReFuNet outperformed existing registration methods (VoxelMorph and SIFT) and fusion methods (IFCNN and DenseFuse), achieving 84.34% accuracy and 87.16% G-mean on the FY test dataset, and 82.88% accuracy and 85.54% G-mean on the HMW test dataset. These results showed significant improvements, particularly in unaligned data scenarios, highlighting the potential for real-world typhoon monitoring and forecasting.
Citation: Zongsheng Zheng, Jia Du, Yuewei Zhang, Xulong Wang. CoReFuNet: A coarse-to-fine registration and fusion network for typhoon intensity classification using multimodal satellite imagery[J]. Electronic Research Archive, 2025, 33(4): 1875-1901. doi: 10.3934/era.2025085
[1] | Maya Briani, Benedetto Piccoli . Fluvial to torrential phase transition in open canals. Networks and Heterogeneous Media, 2018, 13(4): 663-690. doi: 10.3934/nhm.2018030 |
[2] | Xavier Litrico, Vincent Fromion . Modal decomposition of linearized open channel flow. Networks and Heterogeneous Media, 2009, 4(2): 325-357. doi: 10.3934/nhm.2009.4.325 |
[3] | Graziano Guerra, Michael Herty, Francesca Marcellini . Modeling and analysis of pooled stepped chutes. Networks and Heterogeneous Media, 2011, 6(4): 665-679. doi: 10.3934/nhm.2011.6.665 |
[4] | Didier Georges . Infinite-dimensional nonlinear predictive control design for open-channel hydraulic systems. Networks and Heterogeneous Media, 2009, 4(2): 267-285. doi: 10.3934/nhm.2009.4.267 |
[5] | Rudy R. Negenborn, Peter-Jules van Overloop, Tamás Keviczky, Bart De Schutter . Distributed model predictive control of irrigation canals. Networks and Heterogeneous Media, 2009, 4(2): 359-380. doi: 10.3934/nhm.2009.4.359 |
[6] | Nora Aïssiouene, Marie-Odile Bristeau, Edwige Godlewski, Jacques Sainte-Marie . A combined finite volume - finite element scheme for a dispersive shallow water system. Networks and Heterogeneous Media, 2016, 11(1): 1-27. doi: 10.3934/nhm.2016.11.1 |
[7] | Jacek Banasiak, Proscovia Namayanja . Asymptotic behaviour of flows on reducible networks. Networks and Heterogeneous Media, 2014, 9(2): 197-216. doi: 10.3934/nhm.2014.9.197 |
[8] | Avner Friedman . PDE problems arising in mathematical biology. Networks and Heterogeneous Media, 2012, 7(4): 691-703. doi: 10.3934/nhm.2012.7.691 |
[9] | Ye Sun, Daniel B. Work . Error bounds for Kalman filters on traffic networks. Networks and Heterogeneous Media, 2018, 13(2): 261-295. doi: 10.3934/nhm.2018012 |
[10] | Gildas Besançon, Didier Georges, Zohra Benayache . Towards nonlinear delay-based control for convection-like distributed systems: The example of water flow control in open channel systems. Networks and Heterogeneous Media, 2009, 4(2): 211-221. doi: 10.3934/nhm.2009.4.211 |
Typhoons cause significant damage to coastal and inland areas, making the accurate classification of typhoon cloud images essential for effective monitoring and forecasting. While integrating multimodal data from different satellites can improve classification accuracy, existing methods often rely on aligned images and fail to account for radiometric and structural differences, leading to performance degradation during image fusion. In registration methods designed to address this issue, two-stage approaches inaccurately estimate deformation fields, while one-stage methods typically overlook radiometric differences between typhoon cloud images. Additionally, fusion methods suffer from inherent noise accumulation and insufficient cross-modal feature utilization due to cascaded structures. To address these issues, this study proposed a coarse-to-fine registration and fusion network (CoReFuNet) that integrated a one-stage registration module with a cross-modal fusion module for multimodal typhoon cloud image classification. The registration module adopted a one-stage coarse-to-fine strategy, using cross-modal style alignment to address radiometric difference and global spatial registration by affine transformations to resolve positional differences. Bidirectional local feature refinement (BLFR) then ensured precise adjustments, facilitating fine registration by evaluating feature points in each image. Following registration, the fusion module employed a dual-branch alternating enhancement (DAE) approach, which reduced noise by learning cross-modal mapping relationships and applying feedback-based adjustments. Additionally, a cross-modal feature interaction (CMFI) module merged low-level, high-level, intra-modal, and intermodal features through a residual structure, minimizing modality differences and maximizing feature utilization. Experiments on the FY-HMW (Feng Yun-Himawari) dataset, constructed using data from the Feng Yun and Himawari satellites, showed that the CoReFuNet outperformed existing registration methods (VoxelMorph and SIFT) and fusion methods (IFCNN and DenseFuse), achieving 84.34% accuracy and 87.16% G-mean on the FY test dataset, and 82.88% accuracy and 85.54% G-mean on the HMW test dataset. These results showed significant improvements, particularly in unaligned data scenarios, highlighting the potential for real-world typhoon monitoring and forecasting.
The dynamics of network flows is usually modelled by systems of Partial Differential Equations (briefly PDEs), most of time balance laws. The dynamics is defined on a topological graph with evolution on arcs given by system of PDEs, while additional conditions must be assigned at network nodes, e.g. conservation of mass and momentum. There is a large literature devoted to these problems and we refer to [5] for a extensive survey and for additional references.
In particular, here we focus on water flows on a oriented network of open canals and the model given by Saint-Venant or shallow water equations. The latter form a non linear system of balance laws composed by a mass and momentum balance laws. In water management problems, these equations are often used as a fundamental tool to describe the dynamics of canals and rivers, see [1] and papers in same volume, and various control techniques were proposed, see [2,3,17,15,19,23,26] and references therein. Moreover, the need of dynamic models in water management is well documented, see [25]. The shallow water system is hyperbolic (except when water mass vanishes) and has two genuinely nonlinear characteristic fields. Moreover, it exhibits two regimes: fluvial or sub-critical, when one eigenvalue is negative and one positive, and torrential or super-critical, when both eigenvalues are positive. This is captured by the so called Froude number, see (1). For a complete description of the physics of the problem one needs to supply the equations with conditions at nodes, which represent junctions. The junction conditions are originally derived by engineers in the modeling of the dynamic of canals and rivers. The first and most natural condition is the conservation of water mass which is expressed as the equality between the sum of fluxes from the incoming canals and that from outgoing ones. One single condition is not sufficient to isolate a unique solution, thus different additional condition were proposed in the literature. Physical reasons motivate different choices of conditions, among which the equality of water levels, of energy levels and conservation of energy. For the assessment of coupling conditions on canals networks and for more details on the existence of solutions in the case of subcritical flows, we refer the reader to [9,18,16,20,14,22,24]. For discussion on supercritical flow regimes, we refer the reader to [18] and references there in.
Then, to construct solutions one may resort to the concept of Riemann solver at a junction, see [13]. A Riemann solver at a junction is a map assigning solutions to initial data which are constant on each arc. Alternatively one may assign boundary conditions on each arc, but, due to the nonlinearity of equations, one has to make sure that boundary values are attained. This amounts to look for solutions with waves having negative speed on incoming channels and positive on outgoing ones: in other words waves do not enter the junction. A Riemann solver with such characteristics is called consistent, see also [12].
In this paper we are interested in transitions between different flow regimes, when the transition occurs at a junction of a canals network. We assume to have incoming canals which end at the junction and outgoing canals which start at the junction. Thus we formulate a left-half Riemann problem for incoming canals and a right-half Riemann problem for outgoing canals to define the region of admissible states such that waves do not propagate into the junction. This corresponds to identify the regions where Riemann solvers can take values in order to be consistent. Such regions are enclosed by the Lax curves (and inverted Lax curves) and the regime change curves. To help the geometric intuition, we developed pictures showing such curves and the regions they enclose.
The definitions described above and given in Section 4, are the necessary basis for an analysis on a complex network. Due to the complexity of the problem, we consider as case study the specific case of two identical canals interconnected at a junction (simple junction). We start focusing on conservation of water through the junction and equal height as coupling conditions. It is typically expected the downstream flow to be more regular, thus we consider three cases: fluvial to fluvial, torrential to fluvial and torrential to torrential. In the fluvial to fluvial case there exists a unique solution. However such solution may be different than the solution to the same Riemann problem inside a canal (without the junction) and may exhibit the appearance of a torrential regime. The torrential to fluvial case is more delicate to examine. Three different cases may happen: the solution propagates the fluvial regime upstream, the solution propagates the torrential regime downstream or no solution exits. Finally, in the torrential to torrential case, if the solution exists then it is torrential.
To illustrate the achieved results we perform simulations using a Runge-Kutta Discontinuous Galerkin scheme [6]. The RKDG method is an efficient, effective and compact numerical approach for simulations of water flow in open canals. Specifically, it is a high-order scheme and compact in the sense that the solution on one computational cell depends only on direct neighboring cells via numerical fluxes, thus allowing for easy handling the numerical boundary condition at junctions. In the first example we show a simulation where an upstream torrential regime is formed starting from special fluvial to fluvial conditions. The second example shows how a torrential regime may propagate downstream.
We conclude by discussing the possible solutions if the water height condition is replaced by the equal energy condition.
The paper is organized as follows: in Section 2, we present the model starting from the one-dimensional shallow water equations. In Section 3, we give useful notations and preliminary results that allow to determine the admissible states for the half-Riemann problems discussed in the following Section 4. In Section 5 we study possible solutions at a simple junction for different flow regimes and different junction conditions. Finally, in Section 6 we illustrate the results of the previous section with a couple of numerical tests.
The most common and interesting method of classifying open-channel flows is by dimensionless Froude number, which for a rectangular or very wide channel is given by the formula:
Fr=|v|√gh, | (1) |
where
●
●
●
The Froude-number denominator
We are interested in the transition between different flow regimes when it occurs at a junction of a canals network. On each canal the dynamics of water flow is described by the following system of one-dimensional shallow water equations
(hhv)t+(hvhv2+12gh2)x=0. | (2) |
The quantity
∂tu+∂xf(u)=0, | (3) |
where
u=(hhv),f(u)=(hvhv2+12gh2). | (4) |
For smooth solutions, these equations can be rewritten in quasi-linear form
∂tu+f′(u)∂xu=0, | (5) |
where the Jacobian matrix
f′(u)=(01−v2+gh2v). | (6) |
The eigenvalues of
λ1=v−√gh,λ2=v+√gh, | (7) |
with corresponding eigenvectors
r1=(1λ1),r2=(1λ2). | (8) |
The shallow water equations are strictly hyperbolic away from
Solutions to systems of conservation laws are usually constructed via Glimm scheme of wave-front tracking [10,21]. The latter is based on the solution to Rieman problems:
{∂tu+∂xf(u)=0,u(x,0)={ul if x<0,ur if x>0. | (9) |
Here
(R) Centered Rarefaction Waves. Assume
u(x,t)={u− for x<λi(u−)t,Ri(x/t;u−) for λi(u−)t≤x≤λi(u+)t,u+ for x>λi(u+)t, |
where, for the 1-family
R1(ξ;u−):=(19(v−+2√h−−ξ)2127(v−+2√h−+2ξ)(v−+2√h−−ξ)2) |
for
R2(ξ;u−):=(19(−v−+2√h−−ξ)2127(v−2√h−+2ξ)(−v−+2√h−−ξ)2) |
for
(S) Shocks. Assume that the state
u(x,t)={u− if x<λtu+ if x>λt |
provides a piecewise constant solution to the Riemann problem. For strictly hyperbolic systems, where the eigenvalues are distinct, we have that
λi(u+)<λi(u−,u+)<λi(u−),λi(u−,u+)=q+−q−h+−h−. |
To determine a solution for problems on a network, we need to analyze in detail the shape of shocks and rarefaction curves and, more generally, of Lax curves (which are formed by joining shocks and rarefaction ones, see [4]). We start fixing notations and illustrating the shapes of curves.
For a given point
for h<h0,v=R1(h0,v0;h)=v0−2(√gh−√gh0);for h>h0,v=S1(h0,v0;h)=v0−(h−h0)√gh+h02hh0;for h>h0,v=R2(h0,v0;h)=v0−2(√gh0−√gh);for h<h0,v=S2(h0,v0;h)=v0−(h0−h)√gh+h02hh0. | (10) |
Moreover, we define the inverse curves:
for h>h0,v=R−11(h0,v0;h)=v0+2(√gh0−√gh);for h<h0,v=S−11(h0,v0;h)=v0+(h0−h)√gh+h02hh0. | (11) |
Similarly, we set:
for h<h0,v=R−12(h0,v0;h)=v0+2(√gh−√gh0);for h>h0,v=S−12(h0,v0;h)=v0+(h−h0)√gh+h02hh0. | (12) |
We will also consider the regime transition curves: the 1-critical curve is given by
v=C+(h)=√gh | (13) |
and the 2-critical curve by
v=C−(h)=−√gh. | (14) |
In Figure 1 we illustrate the shape of these curves.
To construct a solution to a Riemann problem
ϕl(h):=R1(hl,vl;h)∪S1(hl,vl;h). | (15) |
For the right state
ϕr(h):=R−12(hr,vr;h)∪S−12(hr,vr;h). | (16) |
Remark 1. The Riemann problem for shallow water equations 2 with left state
vl+2√ghl≥vr−2√ghr. | (17) |
When working with
˜ϕl(h)=hϕl(h),˜ϕr(h)=hϕr(h)and˜C+(h)=hC+(h),˜C−(h)=hC−(h). |
Moreover, for a given value
Fi=vi√ghi,or˜Fi=qihi√ghi. | (18) |
In this subsection we study in detail the properties of the function
˜ϕl(h)={h(vl+2√ghl−2√gh),0<h≤hl,h(vl−√g2hl(h−hl)√h+hlh),h>hl |
with
limh→0+˜ϕl(h)=0andlimh→+∞˜ϕl(h)=−∞. |
By computing its first and second derivatives,
˜ϕ′l(h)={vl+2√ghl−3√gh,0<h≤hl,vl−√g2hl(4h2+hlh−h2l2√h(h+hl)),h>hl |
and
˜ϕ′′l(h)={−32√gh,0<h≤hl,−√g2hl(8h3+12hlh+3h2lh+h3l4h(h+hl)√h(h+hl)),h>hl, |
we can conclude that
˜ϕ′l(0)=vl+2√ghl and limh→+∞˜ϕl(h)=−∞ |
we investigate two different cases:
Case 1. If
Case 2. If
Case 2.1. For
h+l,R=19g(vl+2√ghl)2, | (19) |
and we have that the maximum point
˜ϕl(h)=0⇔h=14g(vl+2√ghl)2. |
Notice that for
h−l,R=1g(vl+2√ghl)2; | (20) |
while for
vl−(h−l,S−hl)√gh−l,S+hl2hlh−l,S+√gh−l,S=0; | (21) |
Case 2.2. For
h+l,Ssuch thatvl−(h+l,S−hl)√gh+l,S+hl2hlh+l,S−√gh+l,S=0; | (22) |
and the curve
Notice that for
ql=˜ϕl(h)=hS1(hl,vl;h) |
has two solutions:
h=hlandh=h∗l=hl2(−1+√1+8F2l). | (23) |
Moreover, for
(q2lg)13<hl2(−1+√1+8F2l),forFl>1. |
Indeed, using the relation
Here we study the properties of the function
˜ϕr(h)={h(vr−2√ghr+2√gh),0<h≤hr,h(vr+√g2hr(h−hr)√h+hrh),h>hr. |
By straightforward computations we get its derivatives:
˜ϕ′r(h)={vr−2√ghr+3√gh,0<h≤hr,vr+√g2hr(4h2+hrh−h2r2√h(h+hr)),h>hr. |
˜ϕ′′r(h)={32√gh,0<h≤hr,√g2hr(8h3+12hrh+3h2rh+h3r4h(h+hr)√h(h+hr)),h>hr. |
Then,
˜ϕ′r(0)=vr−2√ghr and limh→+∞˜ϕr(h)=+∞ |
we investigate two different cases:
Case 1. If
Case 2. If
Case 2.1. For
h−r,R=19g(−vr+2√ghr)2 | (24) |
and we have that the minimum point
˜ϕr(h)=0⇔h=14g(−vr+2√ghr)2. |
For
h+r,R=1g(−vr+2√ghr)2, | (25) |
while for
vr+(h+r,S−hr)√gh+r,S+hr2hrh+r,S+√gh+r,S=0. | (26) |
Case 2.2. For
h−r,S such that vr+(hr−h+r,S)√gh−r,S+hr2hrh−r,S+√gh−r,S=0 |
and curve
Notice that the equation
qr=˜ϕr(h)=hS−12(hr,vr;h),h≥hr, |
has two solutions:
h=hrandh=h∗r=hr2(−1+√1+8F2r). | (27) |
Moreover, for
[The case of an incoming canal] We fix a left state and we look for the right states attainable by waves of non-positive speed.
Fix
{∂tu+∂xf(u)=0,u(x,0)={ul if x<0ˆu if x>0 | (28) |
contains only waves with non-positive speed. We distinguish three cases:
● Case A: the left state
● Case B: the left state
● Case C: the left state
For this case we refer to Figure 4.We identify the set
IA1={(ˆh,ˆq):h+l,R≤ˆh≤h−l,S,ˆq=˜ϕl(ˆh)}, | (29) |
where the points
IA2={(ˆh,ˆq):0<ˆh≤h−l,S, ˆq≤ˆhS2(h−l,S,C−(h−l,S);ˆh)}⋃{(ˆh,ˆq):ˆh>h−l,S, ˆq≤˜C−(ˆh)}. | (30) |
The last region
λ(um,ˆu)=qm−ˆqhm−ˆh≤0. |
To define this region, we have to look for values
qm−q=(hm−h)(vm+√g2hm√h(h+hm))≤0,h<hm. |
This inequality is verified for
h∗m=hm2(−1+√1+8F2m). |
We obtain (see Figure 4),
IA3={(ˆh,ˆq): for all (hm,qm) which vary on ˜ϕl such that −1≤˜Fm<0, 0<ˆh≤h∗m, ˆq=ˆhS2(hm,vm;ˆh)}. | (31) |
For this case we refer to Figure 5. It is always possible to connect the left value
h∗l=12(−1+√1+8F2l)hl |
as previously computed in 23. Moreover, as previously observed at the end of subsection 3.0.1 the value
NB(ul)=NA(ul)∖{ˆu=(ˆh,ˆq):h+l,S≤ˆh≤h∗l, ˆq=˜ϕl(ˆh)}, | (32) |
where
For this case we refer to Figure 6. We have that: if
NC(ul)={(ˆh,ˆq):ˆh>0, ˆq<˜C−(ˆh)}; |
otherwise if
NC(ul)={(ˆh,ˆq):0<ˆh≤h−l,R, q<ˆhS2(h−l,R,v−l,R;ˆh)}⋃{(ˆh,ˆq):ˆh>h−l,R, ˆq<˜C−(ˆh)}. | (33) |
[The case of an outgoing canal] We fix a right state and we look for the left states attainable by waves of non-negative speed. For sake of space the figures illustrating these cases will be postponed to the Appendix.
Fix
{∂tu+∂xf(u)=0,u(x,0)={˜u if x<0ur if x>0 | (34) |
contains only waves with non-negative speed. As in the previous case we identify three cases:
● Case A: the right value
● Case B: the right value
● Case C: the right value
For this case we refer to Figure 12 in the Appendix. We identify the set
OA1={(˜h,˜q):h−r,R≤˜h≤h+r,S, ˜q=˜ϕr(˜h)}, | (35) |
where the points
The second region is such that
OA2={(˜h,˜q):0<˜h≤h+r,S, ˜q≥˜hS−11(h+r,S,v+r,S;˜h)}⋃{˜h≥h+r,S, ˜q≥˜C+(˜h)}. | (36) |
The third region is defined by the set of all possible left states
λ(um,˜u)=qm−˜qhm−˜h≥0. |
To define this region we have to look for values
OA3={(˜h,˜q): for all (hm,qm) which vary on ˜ϕr such that 0<˜Fm≤1 0<˜h≤h∗m, ˜q=˜hS−11(hm,vm;˜h)}. | (37) |
For this case we refer to Figure 13 in the Appendix. If
PB(ur)={(˜h,˜q): ˜h≥0, ˜q≥˜C+(˜h)}; |
otherwise, if
PB(ur)={(˜h,˜q): 0<˜h≤h+r,R, ˜q≥˜hS−11(u−r,R;˜h)}⋃{(˜h,˜q),˜h>h−r,R, ˜q≥˜C+(˜h)}. | (38) |
For this case we refer to Figure 14 in the Appendix. It is always possible to connect the right value
h∗r=12(−1+√1+8F2r)hr |
as done in 27. Moreover, as previously observed at the end of subsection 3.0.2, the point
PC(ur)=PA(u∗r)∖{(h,q):h−r,S≤h≤h∗r, q=˜ϕr(h)}, | (39) |
where
We consider here as case study a fictitious network formed by two canals intersecting at one single point, which artificially represents the junction. The junction is straight and separates two equal canals, one is the continuation of the other. This simple scenario appears to be like considering a problem for one straight canal, but by adding a fictitious junction we mimic a network and we provide the first analysis necessary for addressing more complicated networks for which the solution strongly depends on the given assumptions at the junction.
We name the canals such that 1 is the incoming canal and 2 is the outgoing ones. We indicate by
A Riemann Problem at a junction is a Cauchy Problem with initial data which are constant on each canal incident at the junction. So, assuming constant initial conditions
qb1=qb2 | (40) |
and equal heights
hb1=hb2. | (41) |
In the following we study the boundary solution
Case A
{ub1∈NA(ul),ub2∈PA(ur),qb1=qb2=qb,hb1=hb2, | (42) |
with
Proposition 1. Under the subcritical condition on
Proof. We distinguish two cases:
Case 1. The two curves
Case 2. The two curves
ub1=ub2=u+l,R. | (43) |
If
ub1=ub2=u−r,R. | (44) |
Remark 2. Notice that the proposed procedure may give a solution which is different from the classical solution of the Riemann problem on a single channel, given by the intersection point of
Case B
{ub1∈NB(ul),ub2∈PA(ur),qb1=qb2=qb,hb1=hb2, | (45) |
with
Proposition 2. System 45 admits a solution if the two regions
Proof. We distinguish two cases:
Case 1. The two curves
Case 2. The two curves
Case 2.1. Referring to Figure 16, the point
Case 2.2. Referring Figure 17, if
Case B
Assuming different conditions at the junction give rise to new possible solutions. In canals network problems, it is usual to couple the conservation of the mass with the conservation of energy at the junctions. The specific energy
E=h+v22g. | (46) |
For a given flow rate, there are usually two states possible for the same specific energy. Studying
h=hc=(q2g)13. | (47) |
Critical depth
In our case, assuming equal energy at the junction gives
v212+gh1=v222+gh2. | (48) |
Moreover, assuming
gh1F212+gh1=gh31F212h22+gh2, |
where
v21=gh1F21 and v22=v21h21h22=gh31F21h22. |
Then, we have two possible solution for the heights values at the junction:
hb1=hb2 (equal heigths) | (49) |
or
hb2hb1=F214(1+√1+8F21). | (50) |
So, for
Remark 3. In the case of a simple junction, the natural assumption (consistent with the dynamic of shallow-water equations) should be to assume the conservation of the momentum. With our notation, the relation 49 or 48 sholud be replaced by the following:
q21h1+12gh21=q22h2+12gh21. | (51) |
By the same reasoning used before in the case of the conservation of energy, from 51 we get
(h2h1)3−(2F21+1)(h2h1)+2F21=0. |
Then, we have again two possible relations for the heights values at the junction:
hb1=hb2 (equal heigths) | (52) |
or
hb2hb1=12(−1+√1+8F21). | (53) |
So again, for
Let us conclude observing that for appropriate values of
In this Section we illustrate the results of Section 5 by means of numerical simulations. We first give a sketch of the adopted numerical procedure and then we focus on two numerical tests which illustrate the regime transitions from fluvial to torrential and viceversa. The latter depend on well chosen initial conditions for Riemann problems at the junction.
We consider again a network formed by two canals intersecting at one single point, which represents the junction. Following [6], we use a high order Runge-Kutta Discontinuous Galerkin scheme to numerically solve system 3 on both canals 1 and 2:
∂tu1+∂xf(u1)=0,for x<0,∂tu2+∂xf(u2)=0,for x<0. | (54) |
The 1D domain of each canal is discretized into cells
∫Cmw(x)∂tUdx=∫Cmf(U)∂xw(x)dx−(ˆfm+12w−m+12−ˆfm−12w+m−12). | (55) |
Terms
u1(x,t)=k∑l=0ˆu1,lm(t)ψlm(x), | (56) |
where
Once the numerical procedure on both canals has been settled, the two systems in 55 have to be coupled with boundary conditions. At the junction the boundary values is settled as follows: at each time step and at each RK stage via the method-of-line approach, we set as left state in 42 (or 45) the approximate solution from canal 1 at the left limit of the junction, i.e.
ul≈Ul=limx→x−M+12U1(x,⋅) |
with
ur≈Ur=limx→x+,212U2(x,⋅), |
with
ˆfM+12≐f(ub1)for the canal 1 ,ˆf12≐f(ub2)for the canal 2 . |
Finally, in our simulations we assume Neumann boundary conditions at the free extremity of the channels.
Applying this numerical procedure, in Figure 10 and 11 we give two examples which illustrate the solution that is obtained in the regime transitions from fluvial to torrential and viceversa. In Figure 10, we assume to have a starting configuration given by the following subcritical constant states:
This paper deals with open canal networks. The interest stems out of applications such as irrigation channels water management. We base our investigations on the well-known Saint-Venant or shallow water equations. Two regimes exist for this hyperbolic system of balance laws: the fluvial, corresponding to eigenvalues with different sign, and the torrential, corresponding to both positive eigenvalues. Most authors focused the attention on designing and analysing network dynamics for the fluvial regime, while here we extend the theory to include regime transitions. After analyzing the Lax curves for incoming and outgoing canals, we provide admissibility conditions for Riemann solvers, describing solutions for constant initial data on each canal. Such analysis allows to define uniquely dynamics according to a set of conditions at junctions, such as conservation of mass, equal water height or equal energy. More precisely, the simple case of one incoming and outgoing canal is treated showing that, already in this simple example, regimes transitions appear naturally at junctions. Our analysis is then visualized by numerical simulations based on Runge-Kutta Discontinuous Galerkin methods.
M. Briani is a member of the INdAM Research group GNCS.
Here we collect additional figures illustrating attainable regions for half-riemann problems and solutions for a simple channel. Figures 12-14 refer to the right-half Riemann problem described in Section 4.2. They show the regions of admissible states such that waves on the outgoing canals do not propagate into the junction, given a right state
Figures 15-18 refer to Section 5 in which we study the possible solutions at a simple junction for different flow regimes, assuming the conservation of mass and equal heights at the junction. Specifically, Figures 15-17 illustrate the possible configurations and their associated solution that may occur during the transition from torrential to fluvial regime. The last Figure 18 shows instead the only possible configuration that admits a solution for the torrential flow regime.
[1] |
X. Sun, Y. Tian, W. Lu, P. Wang, R. Niu, H. Yu, et al., From single-to multi-modal remote sensing imagery interpretation: A survey and taxonomy, Sci. China Inf. Sci., 66 (2023), 140301. https://doi.org/10.1007/s11432-022-3588-0 doi: 10.1007/s11432-022-3588-0
![]() |
[2] |
L. Li, H. Ling, M. Ding, H. Cao, H. Hu, A deep learning semantic template matching framework for remote sensing image registration, ISPRS, 181 (2021), 205–217. https://doi.org/10.1016/j.isprsjprs.2017.12.012 doi: 10.1016/j.isprsjprs.2017.12.012
![]() |
[3] |
R. Hang, Z. Li, P. Ghamisi, D. Hong, G. Xia, Q. Liu, Classification of hyperspectral and LiDAR data using coupled CNNs, IEEE Trans. Geosci. Remote Sens., 58 (2020), 4939–4950. https://doi.org/10.1109/TGRS.2020.2969024 doi: 10.1109/TGRS.2020.2969024
![]() |
[4] | S. Morchhale, V. P. Pauca, R. J. Plemmons, T. C. Torgersen, Classification of pixel-level fused hyperspectral and lidar data using deep convolutional neural networks, in 2016 8th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), (2016), 1–5. https://doi.org/10.1109/WHISPERS.2016.8071715 |
[5] |
S. Deldari, H. Xue, A. Saeed, D. V. Smith, F. D. Salim, Cocoa: Cross modality contrastive learning for sensor data, Proc. ACM Interact. Mobile Wearable Ubiquitous Technol., 6 (2022), 1–28. https://doi.org/10.1145/3550316 doi: 10.1145/3550316
![]() |
[6] |
D. Hong, L. Gao, N. Yokoya, J. Yao, J. Chanussot, Q. Du, et al., More diverse means better: Multimodal deep learning meets remote-sensing imagery classification, IEEE Trans. Geosci. Remote Sens., 59 (2020), 4340–4354. https://doi.org/10.1109/TGRS.2020.3016820 doi: 10.1109/TGRS.2020.3016820
![]() |
[7] |
H. Li, X. Wu, DenseFuse: A fusion approach to infrared and visible images, IEEE Trans. Image Process., 28 (2018), 2614–2623. https://doi.org/10.1109/TIP.2018.2887342 doi: 10.1109/TIP.2018.2887342
![]() |
[8] |
Y. Zhang, Y. Liu, P. Sun, H. Yan, X. Zhao, X. Zhang, IFCNN: A general image fusion framework based on convolutional neural network, Inf. Fusion, 54 (2020), 99–118. https://doi.org/10.1016/J.INFFUS.2019.07.011 doi: 10.1016/J.INFFUS.2019.07.011
![]() |
[9] |
H. Xu, J. Ma, J. Jiang, X. Guo, H. Ling, U2Fusion: A unified unsupervised image fusion network, IEEE Trans. Pattern Anal. Mach. Intell., 44 (2020), 502–518. https://doi.org/10.1109/TPAMI.2020.3012548 doi: 10.1109/TPAMI.2020.3012548
![]() |
[10] |
N. Li, Y. Li, J. Jiao, Multimodal remote sensing image registration based on adaptive multi-scale PIIFD, Multimed. Tools Appl., 83 (2024), 1–13. https://doi.org/10.1007/s11042-024-18756-1 doi: 10.1007/s11042-024-18756-1
![]() |
[11] | H. Xie, J. Qiu, Y. Dai, Y. Yang, C. Cheng, Y. Zhang, SA-DNet: A on-demand semantic object registration network adapting to non-rigid deformation, preprint, arXiv: 2210.09900. https://doi.org/10.48550/arXiv.2210.09900 |
[12] |
R. Feng, H. Shen, J. Bai, X. Li, Advances and opportunities in remote sensing image geometric registration: A systematic review of state-of-the-art approaches and future research directions, IEEE Trans. Geosci. Remote Sens., 9 (2021), 120–142. https://doi.org/10.1109/MGRS.2021.3081763 doi: 10.1109/MGRS.2021.3081763
![]() |
[13] |
W. M. Wells Ⅲ, P. Viola, H. Atsumi, S. Nakajima, R. Kikini., Multi-modal volume registration by maximization of mutual information, Med. Image Anal., 1 (1996), 35–51. https://doi.org/10.1016/s1361-8415(01)80004-9 doi: 10.1016/s1361-8415(01)80004-9
![]() |
[14] |
A. Goshtasby, G. C. Stockman, C. V. Page, A region-based approach to digital image registration with subpixel accuracy, IEEE Trans. Geosci. Remote Sens., 24 (1986), 390–399. https://doi.org/10.1109/TGRS.1986.289597 doi: 10.1109/TGRS.1986.289597
![]() |
[15] | X. He, C. Meile, S. M. Bhandarkar, Multimodal registration of FISH and nanoSIMS images using convolutional neural network models, preprint, arXiv: 2201.05545. https://doi.org/10.48550/arXiv.2201.05545 |
[16] | D. G. Lowe, Object recognition from local scale-invariant features, in Proceedings of the Seventh IEEE International Conference on Computer Vision, 2 (1999), 1150–1157. https://doi.org/10.1109/ICCV.1999.790410 |
[17] |
J. Li, Q. Hu, M. Ai, RIFT: Multi-modal image matching based on radiation-variation insensitive feature transform, IEEE Trans. Image Process., 29 (2019), 3296–3310. https://doi.org/10.1109/TIP.2019.2959244 doi: 10.1109/TIP.2019.2959244
![]() |
[18] |
L. Tang, Y. Deng, Y. Ma, J. Huang, J. Ma, SuperFusion: A versatile image registration and fusion network with semantic awareness, IEEE/CAA J. Autom. Sin., 9 (2022), 2121–2137. https://doi.org/10.1109/JAS.2022.106082 doi: 10.1109/JAS.2022.106082
![]() |
[19] |
B. K. P. Horn, B. G. Schunck, Determining optical flow, Artif. Intell. 17 (1981), 185–203. https://doi.org/10.1016/0004-3702(81)90024-2 doi: 10.1016/0004-3702(81)90024-2
![]() |
[20] |
J. Xiong, Y. Luo, G. Tang, An improved optical flow method for image registration with large-scale movements, Acta Autom. Sin., 34 (2008), 760–764. https://doi.org/10.3724/SP.J.1004.2008.00760 doi: 10.3724/SP.J.1004.2008.00760
![]() |
[21] |
H. Li, J. Zhao, J. Li, Z. Yu, G. Lu, Feature dynamic alignment and refinement for infrared-visible image fusion: Translation robust fusion, Inf. Fusion, 95 (2023), 26–41. https://doi.org/10.1016/j.inffus.2023.02.011 doi: 10.1016/j.inffus.2023.02.011
![]() |
[22] | S. Tang, P. Miao, X. Gao, Y. Zhong, D. Zhu, H. Wen, et al., Point cloud-based registration and image fusion between cardiac SPECT MPI and CTA, preprint, arXiv: 2402.06841. https://doi.org/10.48550/arXiv.2402.06841 |
[23] | Z. Xu, J. Yan, J. Luo, X. Li, J. Jagadeesan, Unsupervised multimodal image registration with adaptative gradient guidance, in ICASSP 2021–2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), (2021), 1225–1229. https://doi.org/10.1109/ICASSP39728.2021.9414320 |
[24] | H. Xu, J. Ma, J. Yuan, Z. Le, W. Liu, RFNet: Unsupervised network for mutually reinforcing multi-modal image registration and fusion, in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, (2022), 19647–19656. https://doi.org/10.1109/CVPR52688.2022.01906 |
[25] |
H. Xu, J. Yuan, J. Ma, Murf: Mutually reinforcing multi-modal image registration and fusion, IEEE Trans. Pattern Anal. Mach. Intell., 45 (2023), 12148–12166. https://doi.org/10.1109/TPAMI.2023.3283682 doi: 10.1109/TPAMI.2023.3283682
![]() |
[26] | D. Wang, J. Liu, X. Fan, R. Liu, Unsupervised misaligned infrared and visible image fusion via cross-modality image generation and registration, preprint, arXiv: 2205.11876. https://doi.org/10.48550/arXiv.2205.11876 |
[27] |
D. Wang, J. Liu, L. Ma, R. Liu, X. Fan, Improving misaligned multi-modality image fusion with one-stage progressive dense registration, IEEE Trans. Circuits Syst. Video Technol., 34 (2024). https://doi.org/10.1109/TCSVT.2024.3412743 doi: 10.1109/TCSVT.2024.3412743
![]() |
[28] | Z. Zhang, H. Li, T. Xu, X. Wu, J. Kittler, BusReF: Infrared-visible images registration and fusion focus on reconstructible area using one set of features, preprint, arXiv: 2401.00285. https://doi.org/10.48550/arXiv.2401.00285 |
[29] |
L. Z. Li, L. Han, M. Ding, H. Cao, Multimodal image fusion framework for end-to-end remote sensing image registration, IEEE Trans. Geosci. Remote Sens., 61, (2023), 1–14. https://doi.org/10.1109/TGRS.2023.3247642 doi: 10.1109/TGRS.2023.3247642
![]() |
[30] |
S. Mai, Y. Zeng, H. Hu, Multimodal information bottleneck: Learning minimal sufficient unimodal and multimodal representations, IEEE Trans. Multimedia, 25 (2022), 4121–4134. https://doi.org/10.1109/TMM.2022.3171679 doi: 10.1109/TMM.2022.3171679
![]() |
[31] |
Q. Wang, Y. Chi, T. Shen, J. Song, Z. Zhang, Y. Zhu, Improving RGB-infrared object detection by reducing cross-modality redundancy, Remote. Sens., 14 (2020). https://doi.org/10.3390/rs14092020 doi: 10.3390/rs14092020
![]() |
[32] |
S. Cui, J. Cao, X. Cong, J. Sheng, Q. Li, T. Liu, et al., Enhancing multimodal entity and relation extraction with variational information bottleneck, IEEE/ACM Trans. Audio Speech Lang. Process., 32 (2024), 1274–1285. https://doi.org/10.1109/TASLP.2023.3345146 doi: 10.1109/TASLP.2023.3345146
![]() |
[33] |
H. Chen, Y. Li, Three-stream attention-aware network for RGB-D salient object detection, IEEE Trans. Image Process., 28 (2019), 2825–2835. https://doi.org/10.1109/TIP.2019.2891104 doi: 10.1109/TIP.2019.2891104
![]() |
[34] |
X. Sun, L. Zhang, H. Yang, T. Wu, Y. Cen, Y. Guo, Enhancement of spectral resolution for remotely sensed multispectral image, IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens., 8 (2014), 2198–2211. https://doi.org/10.1109/JSTARS.2014.2356512 doi: 10.1109/JSTARS.2014.2356512
![]() |
[35] |
S. Malec, D. Rogge, U. Heiden, A. Sanchez-Azofeifa, M. Bachmann, M. Wegmann, Capability of spaceborne hyperspectral EnMAP mission for mapping fractional cover for soil erosion modeling, Remote Sens., 7 (2015), 11776–11800. https://doi.org/10.3390/rs70911776 doi: 10.3390/rs70911776
![]() |
[36] |
N. Yokoya, T. Yairi, A. Iwasaki, Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion, IEEE Trans. Geosci. Remote Sens., 50 (2020), 528–537. https://doi.org/10.1109/TGRS.2011.2161320 doi: 10.1109/TGRS.2011.2161320
![]() |
[37] | L. Mou, X. Zhu, RiFCN: Recurrent network in fully convolutional network for semantic segmentation of high resolution remote sensing images, preprint, arXiv: 1805.02091. https://doi.org/10.48550/arXiv.1805.02091 |
[38] |
H. Ma, X. Yang, R. Fan, W. Han, K. He, L. Wang, Refined water-body types mapping using a water-scene enhancement deep models by fusing optical and SAR images, IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 17 (2024), 17430–17441. https://doi.org/10.1109/JSTARS.2024.3459916 doi: 10.1109/JSTARS.2024.3459916
![]() |
[39] |
D. Hong, N. Yokoya, J. Chanussot, X. Zhu, CoSpace: Common subspace learning from hyperspectral-multispectral correspondences, IEEE Trans. Geosci. Remote Sens., 57 (2019), 4349–4359. https://doi.org/10.1109/TGRS.2018.2890705 doi: 10.1109/TGRS.2018.2890705
![]() |
[40] |
D. Hong, N. Yokoya, N. Ge, J. Chanussot, X. Zhu, Learnable manifold alignment (LeMA): A semi-supervised cross-modality learning framework for land cover and land use classification, ISPRS J. Photogramm. Remote Sens., 147 (2019), 193–205. https://doi.org/10.1016/j.isprsjprs.2018.10.006 doi: 10.1016/j.isprsjprs.2018.10.006
![]() |
[41] |
G. Li, Z. Liu, M. Chen, Z. Bai, W. Lin, H. Ling, Hierarchical alternate interaction network for RGB-D salient object detection, IEEE Trans. Image Process., 30 (2021), 3528–3542. https://doi.org/10.1109/TIP.2021.3062689 doi: 10.1109/TIP.2021.3062689
![]() |
[42] | M. Chen, L. Xing, Y. Wang, Y. Zhang, Enhanced multimodal representation learning with cross-modal kd, in 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2023), 11766–11775. |
[43] |
D. Hong, J. Yao, D. Meng, Z. Xu, J. Chanussot, Multimodal GANs: Toward crossmodal hyperspectral-multispectral image segmentation, IEEE Trans. Geosci. Remote Sens., 59 (2020), 5103–5113. https://doi.org/10.1109/TGRS.2020.3020823 doi: 10.1109/TGRS.2020.3020823
![]() |
[44] |
D. Hong, N. Yokoya, G. Xia, J. Chanussot, X. Zhu, X-ModalNet: A semi-supervised deep cross-modal network for classification of remote sensing data, ISPRS J. Photogramm. Remote Sens., 167 (2020), 12–23. https://doi.org/10.1016/j.isprsjprs.2020.06.014 doi: 10.1016/j.isprsjprs.2020.06.014
![]() |
[45] | J. Huang, X. Huang, J. Yang, Residual enhanced multi-hypergraph neural network, in 2021 IEEE International Conference on Image Processing (ICIP), (2021), 3657–3661. https://doi.org/10.1109/ICIP42928.2021.9506153 |
[46] | N. Xu, W. Mao, A residual merged neutral network for multimodal sentiment analysis, in 2017 IEEE 2nd International Conference on Big Data Analysis (ICBDA), (2017), 6–10. https://doi.org/10.1109/ICBDA.2017.8078794 |
[47] |
K. He, Z. Zhang, Y. Dong, D. Cai, Y. Lu, W. Han, Improving geological remote sensing interpretation via a contextually enhanced multiscale feature fusion network, IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 17 (2024), 6158–6173. https://doi.org/10.1109/JSTARS.2024.3374818 doi: 10.1109/JSTARS.2024.3374818
![]() |
[48] |
A. M. Saxe, Y. Bansal, J. Dapello, M. Advani, A. Kolchinsky, B. D. Tracey, et al., On the information bottleneck theory of deep learning, J. Stat. Mech.: Theory Exp., 2019 (2019), 124020. https://doi.org/10.1088/1742-5468/ab3985 doi: 10.1088/1742-5468/ab3985
![]() |
[49] |
B. Chen, B. Chen, H. Lin, R. L. Elsberry, Estimating tropical cyclone intensity by satellite imagery utilizing convolutional neural networks, Weather Forecast., 34 (2019), 447–465. https://doi.org/10.1007/s13351-024-3186-y doi: 10.1007/s13351-024-3186-y
![]() |
[50] |
J. Lee, J. Im, D. H. Cha, H. Park, S. Sim, Tropical cyclone intensity estimation using multi-dimensional convolutional neural networks from geostationary satellite data, Remote Sens., 12 (2019), 108. https://doi.org/10.3390/rs12010108 doi: 10.3390/rs12010108
![]() |
[51] |
R. Zhang, Q. Liu, R. Hang, Tropical cyclone intensity estimation using two-branch convolutional neural network from infrared and water vapor images, IEEE Trans. Geosci. Remote Sens., 58 (2019), 586–597. https://doi.org/10.1109/TGRS.2019.2938204 doi: 10.1109/TGRS.2019.2938204
![]() |
[52] |
W. Jiang, G. Hu, T. Wu, L. Liu, B. Kim, Y. Xiao, et al., DMANet_KF: Tropical cyclone intensity estimation based on deep learning and Kalman filter from multi-spectral infrared images, IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 16 (2023), 4469–4483. https://doi.org/10.1109/JSTARS.2023.3273232 doi: 10.1109/JSTARS.2023.3273232
![]() |
[53] | M. Brown, S. Süsstrunk, Multi-spectral SIFT for scene category recognition, in CVPR 2011, (2011), 177–184. https://doi.org/10.1109/CVPR.2011.5995637 |
[54] |
L. Tang, H. Zhang, H. Xu, J. Ma, Rethinking the necessity of image fusion in high-level vision tasks: A practical infrared and visible image fusion network based on progressive semantic injection and scene fidelity, Inf. Fusion, 99 (2023), 101870. https://doi.org/10.1016/j.inffus.2023.101870 doi: 10.1016/j.inffus.2023.101870
![]() |
[55] |
G. Balakrishnan, A. Zhao, M. R. Sabuncu, J. Guttag, A. V. Dalca, Voxelmorph: A learning framework for deformable medical image registration, IEEE Trans. Med. Imaging, 38 (2019), 1788–1800. https://doi.org/10.1109/TMI.2019.2897538 doi: 10.1109/TMI.2019.2897538
![]() |
[56] | X. Fan, X. Wang, J. Gao, J. Wang, Z. Luo, R. Liu, Bi-level learning of task-specific decoders for joint registration and one-shot medical image segmentation, in 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), (2024), 11726–11735. https://doi.org/10.1109/CVPR52733.2024.01114 |
[57] | R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, Grad-cam: Visual explanations from deep networks via gradient-based localization, in Proceedings of the IEEE International Conference on Computer Vision, (2017), 618–626. https://doi.org/10.1109/ICCV.2017.74 |
[58] |
V. F. Dvorak, Tropical cyclone intensity analysis and forecasting from satellite imagery, Mon. Weather Rev., 103 (1975), 420–430. https://doi.org/10.1175/1520-0493(1975)103%3C0420:TCIAAF%3E2.0.CO;2 doi: 10.1175/1520-0493(1975)103%3C0420:TCIAAF%3E2.0.CO;2
![]() |
1. | Luis F. Muñoz-Pérez, J.E. Macías-Díaz, An implicit and convergent method for radially symmetric solutions of Higgs' boson equation in the de Sitter space–time, 2021, 165, 01689274, 270, 10.1016/j.apnum.2021.02.018 | |
2. | M. Briani, G. Puppo, M. Ribot, Angle dependence in coupling conditions for shallow water equations at channel junctions, 2022, 108, 08981221, 49, 10.1016/j.camwa.2021.12.021 | |
3. | Aidan Hamilton, Jing-Mei Qiu, Hong Zhang, 2023, Scalable Riemann Solvers with the Discontinuous Galerkin Method for Hyperbolic Network Simulation, 9798400701900, 1, 10.1145/3592979.3593421 |