Pornographic images not only pollute the internet environment, but also potentially harm societal values and the mental health of young people. Therefore, accurately classifying and filtering pornographic images is crucial to maintaining the safety of the online community. In this paper, we propose a novel pornographic image classification model named ViT-DualAtt. The model adopts a CNN-Transformer hierarchical structure, combining the strengths of Convolutional Neural Networks (CNNs) and Transformers to effectively capture and integrate both local and global features, thereby enhancing feature representation accuracy and diversity. Moreover, the model integrates multi-head attention and convolutional block attention mechanisms to further improve classification accuracy. Experiments were conducted using the nsfw_data_scrapper dataset publicly available on GitHub by data scientist Alexander Kim. Our results demonstrated that ViT-DualAtt achieved a classification accuracy of 97.2% ± 0.1% in pornographic image classification tasks, outperforming the current state-of-the-art model (RepVGG-SimAM) by 2.7%. Furthermore, the model achieves a pornographic image miss rate of only 1.6%, significantly reducing the risk of pornographic image dissemination on internet platforms.
Citation: Zengyu Cai, Liusen Xu, Jianwei Zhang, Yuan Feng, Liang Zhu, Fangmei Liu. ViT-DualAtt: An efficient pornographic image classification method based on Vision Transformer with dual attention[J]. Electronic Research Archive, 2024, 32(12): 6698-6716. doi: 10.3934/era.2024313
[1] | Jiang Li, Xiaohui Liu, Chunjin Wei . The impact of fear factor and self-defence on the dynamics of predator-prey model with digestion delay. Mathematical Biosciences and Engineering, 2021, 18(5): 5478-5504. doi: 10.3934/mbe.2021277 |
[2] | Shunyi Li . Hopf bifurcation, stability switches and chaos in a prey-predator system with three stage structure and two time delays. Mathematical Biosciences and Engineering, 2019, 16(6): 6934-6961. doi: 10.3934/mbe.2019348 |
[3] | Fang Liu, Yanfei Du . Spatiotemporal dynamics of a diffusive predator-prey model with delay and Allee effect in predator. Mathematical Biosciences and Engineering, 2023, 20(11): 19372-19400. doi: 10.3934/mbe.2023857 |
[4] | Kawkab Al Amri, Qamar J. A Khan, David Greenhalgh . Combined impact of fear and Allee effect in predator-prey interaction models on their growth. Mathematical Biosciences and Engineering, 2024, 21(10): 7211-7252. doi: 10.3934/mbe.2024319 |
[5] | Honghua Bin, Daifeng Duan, Junjie Wei . Bifurcation analysis of a reaction-diffusion-advection predator-prey system with delay. Mathematical Biosciences and Engineering, 2023, 20(7): 12194-12210. doi: 10.3934/mbe.2023543 |
[6] | Xin-You Meng, Yu-Qian Wu . Bifurcation analysis in a singular Beddington-DeAngelis predator-prey model with two delays and nonlinear predator harvesting. Mathematical Biosciences and Engineering, 2019, 16(4): 2668-2696. doi: 10.3934/mbe.2019133 |
[7] | Yuanfu Shao . Bifurcations of a delayed predator-prey system with fear, refuge for prey and additional food for predator. Mathematical Biosciences and Engineering, 2023, 20(4): 7429-7452. doi: 10.3934/mbe.2023322 |
[8] | Kalyan Manna, Malay Banerjee . Stability of Hopf-bifurcating limit cycles in a diffusion-driven prey-predator system with Allee effect and time delay. Mathematical Biosciences and Engineering, 2019, 16(4): 2411-2446. doi: 10.3934/mbe.2019121 |
[9] | Ting Yu, Qinglong Wang, Shuqi Zhai . Exploration on dynamics in a ratio-dependent predator-prey bioeconomic model with time delay and additional food supply. Mathematical Biosciences and Engineering, 2023, 20(8): 15094-15119. doi: 10.3934/mbe.2023676 |
[10] | Guangxun Sun, Binxiang Dai . Stability and bifurcation of a delayed diffusive predator-prey system with food-limited and nonlinear harvesting. Mathematical Biosciences and Engineering, 2020, 17(4): 3520-3552. doi: 10.3934/mbe.2020199 |
Pornographic images not only pollute the internet environment, but also potentially harm societal values and the mental health of young people. Therefore, accurately classifying and filtering pornographic images is crucial to maintaining the safety of the online community. In this paper, we propose a novel pornographic image classification model named ViT-DualAtt. The model adopts a CNN-Transformer hierarchical structure, combining the strengths of Convolutional Neural Networks (CNNs) and Transformers to effectively capture and integrate both local and global features, thereby enhancing feature representation accuracy and diversity. Moreover, the model integrates multi-head attention and convolutional block attention mechanisms to further improve classification accuracy. Experiments were conducted using the nsfw_data_scrapper dataset publicly available on GitHub by data scientist Alexander Kim. Our results demonstrated that ViT-DualAtt achieved a classification accuracy of 97.2% ± 0.1% in pornographic image classification tasks, outperforming the current state-of-the-art model (RepVGG-SimAM) by 2.7%. Furthermore, the model achieves a pornographic image miss rate of only 1.6%, significantly reducing the risk of pornographic image dissemination on internet platforms.
Dynamics of prey-predator systems has been an important research theme in ecology. It is well known that prey-predator interaction is not only one of the basic interspecies relations for ecological and social systems, but also the basic block of more complicated food chains, food webs, and biochemical network structures [21]. In the biological world, there are many examples of precesses that involve significant delays that cannot be ignored [13]. Examples include the time between fertilization and birth in the case of sexual reproduction, the time between initiation of cellular division and effective division in the case of mitosis, and the time required for digestion in the case of consumption of nutrient and its conversion to viable biomass [1]. Time delay plays a vital role in population dynamics of prey-predator, which has been recognized to contribute critically to the stable or unstable outcomes of prey population due to predation [10]. Population systems are exposed to the impact of a large number of random unpredictable factors. Then, environmental noise component is another important factor that affects prey-predator dynamics and, it related to climate, geographical distribution, geological features, disaster, human intervention, and other environmental factors [14]. Thus, a realistic prey-predator model should include both time delays and environmental noise.
For example, it is reasonable to assume that the death of prey is instantaneous when attacked by their predator but its contribution to the growth of predator population must be delayed by some time delay [25]. Kuang [10] mentioned that animals must take time to digest their food before further activities and responses taking place. Samanta [26] argued that the effect of time delay due to the time required in going from the egg stage to the adult stage, gestation period etc, has to be taken into account. Three different ways of incorporating a constant time delay into the prey-predator models were presented by Martin and Ruan [18]. In the first case, a time delay appears in the prey specific growth term, which is based on the assumption that in the absence of predators prey satisfies the Hutchinson's equation
Since the nature is full of uncertainty and random phenomena, the natural growth of species often does not follow strictly deterministic laws but rather oscillate randomly about some average, so that the population density never attains a fixed value with the advancement of time rather exhibits continuous oscillation around some average values [14]. The basic mechanism and factors of population growth like resources and vital rates—birth, death, immigration and emigration, change non-deterministically due to continuous fluctuations in the environment (e.g. variation in intensity of sunlight, temperature, water level, etc.) [20]. It is necessary and important to consider the corresponding stochastic population model, i.e., the effects of environmental noise, which undeniably arise from either environmental variability or internal species. For example, in [3], the authors extended a classical epidemic model from a deterministic framework to a stochastic differential equation one through introducing random fluctuations. Among the various ways of constructing a stochastic model systems for a given deterministic system, Cai et al. [4] propose a stochastic version of the epidemic model with nonlinear incident rate. Mao et al. [16] revealed that given population systems are subject to environmental noise, it can suppress a potential population explosion. That is to say, different structures of environmental noise may have different effects on the population systems. In our work, we would include noise processes in both prey and predator.
As we mentioned earlier, a realistic prey-predator model should include both time delays and environmental noise. There is a considerate amount of work done by many scholars, for instance [17,2,22,11,23,27,6,12,28,24,5]. Of them, Mao et al. considered the effects of environmental noise on the delay Lotka-Volterra in [17]. Vasilova [27] studied a stochastic Gilpin-Ayala predator-prey model with time-dependent delay. He established sufficient conditions for the existence of a global positive solution of the considered system, and proved stochastically ultimate boundedness, the long-time behavior of trajectories and extinction of species. In [6], Han et al. investigated two-species Lotka-Volterra delayed stochastic predator-prey system, with and without pollution, respectively. They revealed that there exists a unique nonnegative solution in each system that is permanent in time average under certain conditions. The convergence of the distributions of the solutions of a stochastic two-predator one-prey model with time delay was considered by Liu, Bai and Jin et al. [12]. Wang et al. [28] investigated the dynamics of a stochastic FIV model with seasonality analytically and numerically and, established sufficient criteria for extinction and weak persistence of the disease in the mean.
Motivated by the works of [17,27,12], we propose to study a delayed Harrison-type predator-prey model with noise terms that is lack in literature. We aim to explore and address the following questions through our analytic and numerical results via comparisons to the corresponding ODE model:
1. What are the dynamical effects of time delay in the functional response term of the predator growth equation on the Harrison-type predator-prey model?
2. What are the joint effects of time delay and noise on the Harrison-type predator-prey model?
The remaining of this paper is organized as follows. In Section 2, we derive our model and provide a summary on the dynamics of the corresponding ODE model. And we analyze the stability of a positive equilibrium under the effect of time delay or noise or both of them, and provide sufficient conditions that guarantee the existence of Hopf bifurcation at a positive equilibrium. In Section 3, numerical simulations are given to verify the theoretic analysis. In Section 4, the paper provides a brief conclusion and remarks. All proofs of our theoretical results are presented in the last section.
The general predator-prey model can be expressed by the model of nonlinear ordinary differential equations [7]:
{dNdt=u(N)−f(N)g(P),dPdt=αf(N)g(P)−v(P), | (1) |
where
f(N)=cN,g(P)=PmP+1,c,m>0, |
and the function
{dNdt=N(1−N)−cNPmP+1,dPdt=P(−d+bNmP+1), | (2) |
where
In this paper, we will study the effects of fluctuating environment on the dynamical behaviors of the Harrison-type predation model (2) with time delay
{dN(t)=N(t)(1−N(t)−cP(t)mP(t)+1)dt+σ1N(t)dB(t),dP(t)=P(t)(−d+bN(t−τ)mP(t−τ)+1)dt+σ2P(t)dB(t), | (3) |
where
Since our proposed model (3) is the population model, we restrict the state space being the closed first quadrant in the
Theorem 2.1. The positive quadrant
Before further studying the dynamics of our full model (3), we provide the following results regarding its corresponding ODE model (2) and the following delayed model
{dN(t)dt=N(t)(1−N(t))−cN(t)P(t)mP(t)+1,dP(t)dt=P(t)(−d+bN(t−τ)mP(t−τ)+1). | (4) |
Model (3) has the trivial constant steady state
Ni=b(m−c)±√4bcdm+b2(m−c)22bm,Pi=bNi−ddm,i=1,2,N1<N2, |
where equilibrium point
Theorem 2.2(Dynamics of the ODE model). The ODE model (2) always has two boundary equilibria: the extinction equilibrium
1. If
2. If
The existence and stability conditions of the equilibrium of the ODE model (2) is listed in Table 1.
Equilibrium | Existence Condition | Stability Condition |
Always exists | Always saddle | |
Always exists | Sink if Saddle if | |
Always sink |
The proof of Theorem 2.2 is rather standard, and detailed investigations relating to the ODE model may be found in Ref. [29], hence is omitted.
Remark 1. We would like to point out that Wang and Cai [29] provided similar results of the ODE model (2). For the ODE model (2), some preliminary results are presented, including dissipativeness, boundedness, permanence of the solutions, and the equilibria stability analysis of the model. Model (2) has two boundary equilibria
For model (3), when
Theorem 2.3. For the DDE model (4),
(i) The nonnegative solution
lim supt→∞P(t)≤(b−d)exp(τ)dm |
for
(ii) If the condition
Remark 2. In the case of
Theorem 2.4. If the condition
b(m+exp(τ)−√(m+exp(τ))2−4mcexp(τ))2m<d<min{b,b(m+exp(τ)+√(m+exp(τ))2−4mcexp(τ))2m} | (5) |
with
Remark 3. Biologically, Theorem 2.4 means that all the species of the DDE model (4) are survivor under
Let
{dN(t)dt=(N∗+N(t))(1−(N∗+N(t)))−c(N∗+N(t))(P∗+P(t))m(P∗+P(t))+1,dP(t)dt=(P∗+P(t))(−d+b(N∗+N(t−τ))m(P∗+P(t−τ))+1), | (6) |
which can be rewritten as an abstract differential equation in the phase space
dU(t)dt=L(U(t))+G(U(t)), | (7) |
where
L(U(t))=(q1N(t)+q2P(t)q3N(t−τ)+q4P(t−τ)), | (8) |
and
G(U(t))=(−N2(t)−cN(t)P(t)(mP∗+1)2+cmN∗P2(t)(mP∗+1)2bN(t−τ)P(t)mP∗+1−bmN∗P(t)P(t−τ)+P∗N(t−τ)P(t−τ)(mP∗+1)2+bm2N∗P∗P2(t−τ)(mP∗+1)3), | (9) |
here,
q1=−N∗,q2=−cN∗(mP∗+1)2,q3=bP∗mP∗+1,q4=−bmN∗P∗(mP∗+1)2. | (10) |
Referring to [10], we can know that given initial data
The characteristic equation of the linearization model
λ2−q1λ+(q1q4−q2q3−q4λ)e−λτ=0, | (11) |
where
ω0=√−(q21−q24)+√(q21−q24)2+4(q1q4−q2q3)22, | (12) |
and the corresponding critical value of delay
τk=1ω0arcsin(−q4ω30+(q1q4−q2q3)q1ω0(q1q4−q2q3)2+q24ω20)+2kπω0,k=0,1,2,… | (13) |
Hence, we have the following theorem.
Theorem 2.5. For the DDE model (4), if
Remark 4. In the absence of delay
Fig. 2 shows the time-series plots of prey
In the following, we turn our focus on the dynamical behaviors of the SDDE model (3) with both time delay and noise effect. Denote
Theorem 2.6. For any given initial value
Remark 5. Theorem 2.6 shows that the solution of the SDDE model (3) will remain in
The property of Theorem 2.6 makes us continue to discuss how the solution varies in
Definition 2.7. The solution of the SDDE model (3) is said to be stochastically ultimately bounded if for any
lim supt→∞P{|(N(t),P(t))|>χ}<ε. | (14) |
Theorem 2.8. For any
lim supt→∞E|(N(t),P(t))|θ≤C. | (15) |
Theorem 2.9. For any
Remark 6. The above analyses show that under certain conditions, the original ODE model (2) and the associated SDDE model (3) behave similarly in the sense that both have positive solutions which will not explode to infinity in a finite time, and actually, will be ultimately bounded. That is to say, under certain conditions the noise will not spoil these nice properties.
The following result gives the almost surely asymptotic property of model (3).
Theorem 2.10. Assume the same conditions of Theorem 2.6 hold. For any given initial value
lim supt→∞ln|Z(t)|t≤1+d+˘C22ˆσ2a.s. |
where
In this section, numerical simulation results of the SDDE model (3) for different values of
c=0.9,b=0.7,d=0.3,m=0.1 | (16) |
and the initial value is set to
Fig. 3 shows the time-series plots of model (3) only with different noise intensities
Fig. 4 shows the evolution of both the prey and predator species of the SDDE model (3) with different noise intensities
Fig. 5 shows the fluctuation in population densities of the prey and predator species of model (3) for different values of
For the stochastic delayed Harrison-type predator-prey model (3), one of our interesting findings is that our proposed delayed predation model can exhibit Hopf bifurcation. And we prove that the positive equilibrium is asymptotically stable when the delay is less than a certain critical value and unstable when the delay is greater than the critical value. These findings have also been reported in the delayed Harrison-type predator-prey system with diffusive effects under Neumann boundary conditions in [31]. The work of [31] shows that the system has a Hopf bifurcation and, the authors analyze the direction of Hopf bifurcation and the stability of bifurcating periodic solution. In addition, we provide the conditions of dissipative, uniformly bounded, and permanent of our proposed delayed model.
We also provide additional results on the case when the Harrison-type predation model is added by both time delay and environmental noise under certain conditions. Our analysis and numerical simulations suggest that large delay could destabilize the model; and the combination of delay and noise could intensify the periodic instability of the SDDE model (3). Complicated analytical conditions for the above results depend upon certain parametric restrictions involving the parameters of model (3) and magnitudes of time delay and environmental noise. Our analytical results partially provide answers to the two questions proposed in the introduction: What are the dynamical effects of time delay in the functional response term of the predator growth equation on the Harrison-type predator-prey model? What are the joint effects of time delay and noise on the Harrison-type predator-prey model? More specifically, we have follows:
1. The effect of delay induces instability: Theorem 2.5 indicates that the large time delay can induce instability and oscillations via Hopf bifurcation whereas the corresponding model without delay is asymptotically stable at the interior equilibrium. Delay induced instability is shown in Figs. 2(b) and 2(c) and cause the populations to fluctuate. If the time delay
2. The synergistic effects of delay and noise: We illustrate some relevant properties of the corresponding stochastic delayed predator-prey model (3) and reveal the effect of environmental noise on the model. The increase of the noise intensity has a drastic impact on the dynamical behavior of both species with or without the delay effect. Theorem 2.6 shows that under some conditions, the stochastically perturbed SDDE model (3) of the delay model (4) will remain to have a positive stable solution, and this theorem gives a result on the robustness of the positive stable solution. Under the conditions of Theorems 2.8 and 2.9, the property of the ultimate boundedness of the SDDE model (3) will not change no matter the environmental noise is large or small. That is to say, the property of this boundedness is very robust under the noise. In short, under certain conditions, the DDE model (4) and the associated SDDE model (3) behave similarly in the sense that both the prey and predator species have positive solutions which will not explode to infinity in a finite time and, in fact, will be ultimately bounded. That is, under certain conditions environmental noise will not spoil these nice properties. Moreover, Theorem 2.10 shows model (3) has almost surely asymptotic property. By fixing the values of
Proof of Theorem 2.1.
Proof. For all
{N(t)=N(0)exp(∫t0(1−N(s)−cP(s)mP(s)+1)ds+∫t0σ1dB(s)),P(t)=P(0)exp(∫t0(bN(s−τ)mP(s−τ)+1−d)ds+∫t0σ2dB(s)). | (17) |
As
{N(t)=N(0)exp(∫t0(1−N(s)−cP(s)mP(s)+1)ds+σ1B(t))≥N(0)exp(−CT),P(t)=P(0)exp(∫t0(bN(s−τ)mP(s−τ)+1−d)ds+σ2B(t))≥P(0)exp(−CT). | (18) |
Taking the limit, as
N(T)≥N(0)exp(−CT)>0,P(T)≥P(0)exp(−CT)>0, |
which contradicts the fact that either
Proof of Theorem 2.3.
Proof. For our full model (3), when
{dN(t)dt=N(t)(1−N(t))−cN(t)P(t)mP(t)+1,dP(t)dt=P(t)(−d+bN(t−τ)mP(t−τ)+1), | (19) |
with initial conditions
From the predator equation of model (19), we get
dP(t)dt<bP(t), |
hence, for
P(t)≤P(t−τ)exp(bτ), |
which is equivalent, for
P(t−τ)≥P(t)exp(−bτ). |
For any
dP(t)dt<P(t)(εb(mP(t)exp(−ετ)+1)−d), |
which implies by the same arguments use for
lim supt→∞P(t)≤(εb−d)exp(ετ)dm. | (20) |
Conclusion of this Theorem holds by letting
In addition, based on the above analysis, if
Proof of Theorem 2.4.
Proof. From Theorem 2.3, there is a
max{lim supt→∞N(t),lim supt→∞P(t)}≤A. |
Then, we will show that there is a
min{lim inft→∞N(t),lim inft→∞P(t)}≥B. |
From model (4), for any
dN(t)dt>N(t)(1−N(t)−cε(b−d)exp(τ)dm). |
Based on the standard comparison arguments, it follows that
lim inft→∞N(t)≥(1−cε(b−d)exp(τ)dm) |
and letting
lim inft→∞N(t)≥(1−c(b−d)exp(τ)dm). | (21) |
Set
dP(t)dt>P(t)(bC1ε(mP(t−τ)+1)−d). | (22) |
On the one hand, for
dP(t)dt>P(t)(C1bdε(ε(b−d)exp(τ)+d)−d), |
which involves, for
P(t−τ)<P(t)exp(−(C1bdε(ε(b−d)exp(τ)+d)−d)τ). | (23) |
On the other hand, from (22) and (23), for
dP(t)dt>P(t)(bC1εmP(t)exp(−(C1bdε(ε(b−d)exp(τ)+d)−d)τ)+ε−d) |
which yields
lim inft→∞P(t)≥bC1−εdεdmexp(−(d−C1bdε2(b−d)exp(τ)+εd)τ). |
When
lim inft→∞P(t)≥bC1−ddmexp(−(d−C1bd(b−d)exp(τ)+d)τ)=C2. |
Let be
Proof of Theorem 2.5.
Proof. In the absence of delay, i.e.,
Assume that for some
−ω2−iq1ω+(q1q4−q2q3−iq4ω)e−iωτ=0. | (24) |
Separating the real and imaginary parts, we have
{ω2=(q1q4−q2q3)cosωτ−q4ωsinωτ,q1ω=−(q1q4−q2q3)sinωτ−q4ωcosωτ, | (25) |
thus,
sinωτ=−q4ω3+(q1q4−q2q3)q1ω(q1q4−q2q3)2+q24ω2,cosωτ=−q2q3ω2(q1q4−q2q3)2+q24ω2, | (26) |
which leads to the following equation
ω4+(q21−q24)ω2−(q1q4−q2q3)2=0. | (27) |
Solving for
ω2=−(q21−q24)±√(q21−q24)2+4(q1q4−q2q3)22. | (28) |
Thus, Eq. (27) has a unique positive root
ω0=√−(q21−q24)+√(q21−q24)2+4(q1q4−q2q3)22, | (29) |
and substituting this value in (25), then the corresponding critical value of delay
τk=1ω0arcsin(−q4ω30+(q1q4−q2q3)q1ω0(q1q4−q2q3)2+q24ω20)+2kπω0,k=0,1,2,… | (30) |
Let
(dλ(τ)dτ)−1=(2λ−q1)eλτ−(q1q4−q2q3)τ−(1+λτ)q4λ(q1q4−q2q3+q4λ)=(2λ−q1)eλτλ(q1q4−q2q3+q4λ)−q4λ(q1q4−q2q3+q4λ)−τλ, | (31) |
which, together with (25), (26) and (29), leads to
sign{Re(dλdτ)}τ=τk=sign{Re(dλdτ)−1}τ=τk=q21+2ω20ω20(q21+ω20)+q24(q1q4−q2q3)2+q24ω20>0. | (32) |
This implies that all the roots cross the imaginary axis at
Proof of Theorem 2.6.
Proof. From the biological meaning, we only consider the positive solution to the SDDE model (3). Taking the change of variables as
{du(t)=(1−eu(t)−cev(t)mev(t)+1−σ212)dt+σ1dB(t),dv(t)=(−d+beu(t−τ)mev(t−τ)+1−σ212)dt+σ2dB(t), | (33) |
the coefficient of (33) satisfy the local Lipschitz condition, then for any given initial values
1l0<min−τ≤t≤0|Z(t)|≤max−τ≤t≤0|Z(t)|<l0. |
For each integer
tl=inf{t∈[0,te):N(t)∉(1l,l),P(t)∉(1l,l)}, |
where we set
If we can show that
V(Z)=(N−logN−1)+(P−logP−1), | (34) |
which is a nonnegative function on
d[∫tt−τ(N(s)+P(s))ds+V(Z(t))]=[N−N(t−τ)+(N−1)(1−N−cPmP+1)+P−P(t−τ)+(P−1)(bN(t−τ)mP(t−τ)+1−d)+σ21+σ222]dt+σ1(N−1)dB(t)+σ2(P−1)dB(t)≤[54+cm+d−N(t−τ)−(N−32)2+(1−d)P−P(t−τ)+bN(t−τ)P+σ21+σ222]dt+σ1(N−1)dB(t)+σ2(P−1)dB(t)≤[54+cm+d−N(t−τ)−(N−32)2−P(t−τ)−(d−1−b|Z(t−τ)|)P+σ21+σ222]dt+σ1(N−1)dB(t)+σ2(P−1)dB(t)≤Cdt+σ1(N−1)dB(t)+σ2(P−1)dB(t), | (35) |
where
E[∫tl∧Ttl∧T−τ(N(s)+P(s))ds+V(N(tl∧T),P(tl∧T))]≤∫0−τ(N(s)+P(s))ds+V(N(0),P(0))+CT, |
thus,
EV(N(tl∧T),P(tl∧T))≤∫0−τ(N(s)+P(s))ds+V(N(0),P(0))+CT. | (36) |
Set
V(N(tl,μ),P(tl,μ))≥min{l−logl−1,1l+logl−1}. | (37) |
It then follows from (36) that
∫0−τ(N(s)+P(s))ds+V(N(0),P(0))+CT≥E[1Ωl(μ)V(N(tl),P(tl))]≥P{tl≤T}min{l−logl−1,1l+logl−1}, | (38) |
where
Proof of Theorem 2.8.
Proof. Define a function by
V(N,P)=Nθ+Pθ,(N,P)∈R2+. | (39) |
For any
LV(N,P)=θNθ(1−N−cPmP+1)+σ212θ(θ−1)Nθ+θPθ(bN(t−τ)mP(t−τ)+1−d)+σ222θ(θ−1)Pθ, | (40) |
as
LV(N,P)≤θNθ(1−N)−σ212θ(1−θ)Nθ+θPθbN(t−τ)mP(t−τ)+1−σ222θ(1−θ)Pθ≤θNθ−σ212θ(1−θ)Nθ+bθ(b−d)exp(τ)dm|Z(t−τ)|2−σ222θ(1−θ)Pθ=F(N,P)−V(N,P)−eτ|Z(t)|2+bθ(b−d)exp(τ)dm|Z(t−τ)|2 | (41) |
where
LV(N,P)≤C0−V(N,P)−eτ|Z(t)|2+bθ(b−d)exp(τ)dm|Z(t−τ)|2. |
Hence, we obtain
dV(N,P)=LV(N,P)dt+σ1θNθdB(t)+σ2θPθdB(t)≤(C0−V(N,P)−eτ|Z(t)|2+bθ(b−d)exp(τ)dm|Z(t−τ)|2)dt+σ1θNθdB(t)+σ2θPθdB(t). | (42) |
Applying It
d(etV(N,P))=etV(N,P)dt+etdV(N,P)≤et(C0−eτ|Z(t)|2+bθ(b−d)exp(τ)dm|Z(t−τ)|2)dt+etσ1θNθdB(t)+etσ2θPθdB(t). | (43) |
If
etEV(N,P)≤V(N(0),P(0))+C0et−E∫t0es+τ|Z(s)|2ds+bθ(b−d)exp(τ)dmE∫t0es|Z(s−τ)|2ds=V(N(0),P(0))+C0et−E∫t0es+τ|Z(s)|2ds+bθ(b−d)dmE∫t−τ−τes+τ|Z(s)|2ds≤V(N(0),P(0))+C0et+bθ(b−d)dmE∫0−τes+τ|Z(s)|2ds, |
which implies that
lim supt→∞EV(N(t),P(t))≤C0. | (44) |
Since
|(N,P)|θ≤√2θmax{Nθ,Pθ}≤√2θV(N,P), | (45) |
we get
lim supt→∞E|(N(t),P(t))|θ≤√2θlim supt→∞EV(N(t),P(t))≤√2θC0=C(θ). | (46) |
The proof is complete.
Proof of Theorem 2.9.
Proof. The solutions of model (3) will remain in
lim supt→∞E|(N(t),P(t))|θ≤C. | (47) |
Then, we choose a constant
P{Nθ+Pθ>ϵ}≤1ϵE[Nθ+Pθ]≤Cϵ:=ε, |
which implies that
1−ε≤P{Nθ+Pθ<ϵ}≤P{1ϵ≤N(t)+P(t)≤ϵ}. |
Noting
P{1√2ϵ≤N(t)+P(t)√2≤|(N(t),P(t))|≤N(t)+P(t)≤ϵ}≥1−ε. | (48) |
According to the Definition 2.7, the SDDE model (3) is stochastically ultimately bounded.
Proof of Theorem 2.10.
Proof. Define a function by
lnV(Z(t))=lnV(Z(0))+∫t0(NV(Z(s))(1−N−cPmP+1)−σ21N22V2(Z(s)))ds+∫t0(PV(Z(s))(bN(s−τ)mP(s−τ)+1)−d)−σ22P22V2(Z(s)))ds+M1(t)+M2(t), | (49) |
where
⟨Mi(t),Mi(t)⟩=∫t0σ2iZ2iV(Z(s))ds,i=1,2. |
Let
P{sup0≤t≤k(Mi(t)−ε2⟨Mi(t),Mi(t)⟩)>θlnkε}≤k−θ. |
Since the series
Mi(t)≤ε2⟨Mi(t),Mi(t)⟩+2εlnk |
for all
lnV(Z(t))≤lnV(Z(0))+∫t0(NV(Z(s))(1−N−cPmP+1)−(1−ε)σ21N22V2(Z(s)))ds+∫t0(PV(Z(s))(bN(s−τ)mP(s−τ)+1)−d)−(1−ε)σ22P22V2(Z(s)))ds+4εlnk | (50) |
for all
NV(Z(s))(1−N−cPmP+1)≤1+cP,PV(Z(s))(bN(s−τ)mP(s−τ)+1)−d)≤bN(s−τ)+d,σ21N22V2(Z(s))≥σ21c21N2,σ22P22V2(Z(s))≥σ22c22P2 | (51) |
where
lnV(Z(t))≤∫t0(1+cP(s)+bN(s−τ)+d−(1−ε)σ21c21N2−(1−ε)σ22c22P2)ds+lnV(Z(0))+4εlnk≤∫t0(1+cP(s)+bN(s−τ)+d−(1−ε)ˆσ2|Z|22)ds+lnV(Z(0))+4εlnk | (52) |
for all
lnV(Z(t))+(1−2ε)ˆσ24∫t0|Z(s)|2ds≤lnV(Z(0))+4εlnk+∫t0(1+cP(s)+bN(s−τ)+d−ˆσ2|Z|22)ds≤C1+∫t0(1+cP(s)+bN(s)+d−ˆσ2|Z|22)ds+4εlnk, |
where
1+d+cP(t)+bN(t)−ˆσ2|Z|22≤1+d+ˇC|Z|−ˆσ2|Z|22≤1+d+ˇC22ˆσ2=C2, |
thus, if
lnV(Z(t))+(1−2ε)ˆσ24∫t0|Z(s)|2ds≤ˇC+4εlnk+C2t |
for all
1t(lnV(Z(t))+(1−2ε)ˆσ24∫t0|Z(s)|2ds)≤1k−1(ˇC+4εlnk)+C2, |
which implies
lim supt→∞1t(lnV(Z(t))+(1−2ε)ˆσ24∫t0|Z(s)|2ds)≤C2. | (53) |
Letting
lim supt→∞1t(lnV(Z(t))+ˆσ24∫t0|Z(s)|2ds)≤C2a.s. | (54) |
and using
lim supt→∞1t(ln|Z(t)|+ˆσ24∫t0|Z(s)|2ds)≤C2a.s. |
The desired assertion is derived.
This research is partially supported by the National Natural Science Foundation of China (Grant Nos. 11601226, 11426132, 71501094), the Natural Science Foundation of Jiangsu Province of China (Grant No. BK20140927, BK20150961), and the research funds from Nanjing Tech University and Jiangsu Government Scholarship for Overseas Studies. The work is also partially supported by NSF-DMS (1313312&1716802); NSF-IOS/DMS (1558127), and The James S. McDonnell Foundation 21st Century Science Initiative in Studying Complex Systems Scholar Award (UHC Scholar Award 220020472). This project has been partially supported by grants from the National Science Foundation (DMS-1263374 and DMS-1261211), and the Offices of the President and the Provost of Arizona State University.
[1] |
Z. Wang, R. Guo, H. Wang, X. Zhang, A new model for small target adult image recognition, Procedia Comput. Sci., 183 (2021), 557–562. https://doi.org/10.1016/j.procs.2021.02.097 doi: 10.1016/j.procs.2021.02.097
![]() |
[2] |
B. Wang, X. Lv, X. Ma, H. Wang, Application of skin detection based on irregular polygon area boundary constraint on YCbCr and reverse gamma correction, Adv. Mater. Res., 327 (2011), 31–36. https://doi.org/10.4028/www.scientific.net/AMR.327.31 doi: 10.4028/www.scientific.net/AMR.327.31
![]() |
[3] | Z. Zhao, A. Cai, Combining multiple SVM classifiers for adult image recognition, in 2010 2nd IEEE International Conference on Network Infrastructure and Digital Content, IEEE, (2010), 149–153. https://doi.org/10.1109/ICNIDC.2010.5657916 |
[4] |
S. Paheding, A. Saleem, M. F. H. Siddiqui, N. Rawashdeh, A. Essa, A. A. Reyes, Advancing horizons in remote sensing: A comprehensive survey of deep learning models and applications in image classification and beyond, Neural Comput. Appl., 36 (2024), 16727–16767. https://doi.org/10.1007/s00521-024-10165-7 doi: 10.1007/s00521-024-10165-7
![]() |
[5] |
C. Zhao, R. W. Liu, J. Qu, R. Gao, Deep learning-based object detection in maritime unmanned aerial vehicle imagery: Review and experimental comparisons, Eng. Appl. Artif. Intell., 128 (2024), 107513. https://doi.org/10.1016/j.engappai.2023.107513 doi: 10.1016/j.engappai.2023.107513
![]() |
[6] |
R. Shetty, V. S. Bhat, J. Pujari, Content-based medical image retrieval using deep learning-based features and hybrid meta-heuristic optimization, Biomed. Signal Process. Control, 92 (2024), 106069. https://doi.org/10.1016/j.bspc.2024.106069 doi: 10.1016/j.bspc.2024.106069
![]() |
[7] | A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. Gomez, et al., Attention is all you need, preprint, arXiv: 1706.03762. |
[8] |
W. Zhang, G. Chen, P. Zhuang, W. Zhao, L. Zhou, CATNet: Cascaded attention transformer network for marine species image classification, Expert Syst. Appl., 256 (2024), 124932. https://doi.org/10.1016/j.eswa.2024.124932 doi: 10.1016/j.eswa.2024.124932
![]() |
[9] |
M. Ahmad, U. Ghous, M. Usama, M. Mazzara, WaveFormer: Spectral–spatial wavelet transformer for hyperspectral image classification, IEEE Geosci. Remote Sens. Lett., 21 (2024), 1–5. https://doi.org/10.1109/LGRS.2024.3353909 doi: 10.1109/LGRS.2024.3353909
![]() |
[10] |
G. Huang, H. Zhou, X. Ding, R. Zhang, Extreme learning machine for regression and multiclass classification, IEEE Trans. Syst. Man Cybern. Part B Cybern., 42 (2011), 513–529. https://doi.org/10.1109/TSMCB.2011.2168604 doi: 10.1109/TSMCB.2011.2168604
![]() |
[11] | G. Guo, H. Wang, D. Bell, Y. Bi, K. Greer, KNN model-based approach in classification, in On the Move to Meaningful Internet Systems 2003: CoopIS, DOA, and ODBASE, Springer, (2003), 986–996. https://doi.org/10.1007/978-3-540-39964-3_62 |
[12] | H. Zhao, I. Liu, Research on test data generation method of complex event big data processing system based on Bayesian network, Comput. Appl. Res., 35 (2018), 155–158. |
[13] |
M. Jones, J. Rehg, Statistical color models with application to skin detection, Int. J. Comput. Vision, 46 (2002), 81–96. https://doi.org/10.1023/A:1013200319198 doi: 10.1023/A:1013200319198
![]() |
[14] | Y. Lin, H. Tseng, C. Fuh, Pornography detection using support vector machine, in 16th IPPR Conference on Computer Vision, Graphics and Image Processing (CVGIP 2003), (2003), 123–130. |
[15] | L. Lv, C. Zhao, H. Lv, J. Shang, Y. Yang, J. Wang, Pornographic images detection using high-level semantic features, in 2011 Seventh International Conference on Natural Computation, IEEE, (2011), 1015–1018. https://doi.org/10.1109/ICNC.2011.6022151 |
[16] | K. Dong, L. Guo, Q. Fu, An adult image detection algorithm based on Bag-of-Visual-Words and text information, in 2014 10th International Conference on Natural Computation, IEEE, (2014), 556–560. https://doi.org/10.1109/ICNC.2014.6975895 |
[17] | L. Lv, C. Zhao, H. Lv, J. Shang, Y. Yang, J. Wang, Pornographic images detection using high-level semantic features, in 2011 Seventh International Conference on Natural Computation, IEEE, (2011), 1015–1018. https://doi.org/10.1109/ICNC.2011.6022151 |
[18] |
F. Cheng, S. Wang, X. Wang, A. Liew, G. Liu, A global and local context integration DCNN for adult image classification, Pattern Recognit., 96 (2019), 106983. https://doi.org/10.1016/j.patcog.2019.106983 doi: 10.1016/j.patcog.2019.106983
![]() |
[19] |
Z. Cai, X. Hu, Z. Geng, J. Zhang, Y. Feng, An illegal image classification system based on deep residual network and convolutional block attention module, Int. J. Network Secur., 25 (2023), 351–359. https://doi.org/10.6633/IJNS.202303_25(2).18 doi: 10.6633/IJNS.202303_25(2).18
![]() |
[20] |
Z. Cai, X. Qiao, J. Zhang, Y. Feng, X. Hu, N. Jiang, Repvgg-simam: An efficient bad image classification method based on RepVGG with simple parameter-free attention module, Appl. Sci., 13 (2023), 11925. https://doi.org/10.3390/app132111925 doi: 10.3390/app132111925
![]() |
[21] | S. Woo, J. Park, J. Lee, I. Kweon, CBAM: Convolutional Block Attention Module, in Computer Vision – ECCV 2018, Springer, (2018), 3–19. https://doi.org/10.1007/978-3-030-01234-2_1 |
[22] | S. Yu, S. Jin, J. Peng, H. Liu, Y. He, Application of a new deep learning method with CBAM in clothing image classification, in 2021 IEEE International Conference on Emergency Science and Information Technology (ICESIT), IEEE, (2021), 364–368. https://doi.org/10.1109/ICESIT53460.2021.9696783 |
[23] |
J. Liu, H. Qiao, L. Yang, J. Guo, Improved lightweight YOLOv4 foreign object detection method for conveyor belts combined with CBAM, Appl. Sci., 13 (2023), 8465. https://doi.org/10.3390/app13148465 doi: 10.3390/app13148465
![]() |
[24] | J. Devlin, M. Chang, K. Lee, K. Toutanova, BERT: Pre-training of deep bidirectional transformers for language understanding, preprint, arXiv: 1810.04805. |
[25] | A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, et al., An image is worth 16 × 16 words: Transformers for image recognition at scale, preprint, arXiv: 2010.11929. |
[26] |
G. Brauwers, F. Frasincar, A general survey on attention mechanisms in deep learning, IEEE Trans. Knowl. Data Eng., 35 (2021), 3279–3298. https://doi.org/10.1109/TKDE.2021.3126456 doi: 10.1109/TKDE.2021.3126456
![]() |
[27] | 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), IEEE, (2016), 770–778. https://doi.org/10.1109/CVPR.2016.90 |
[28] | K. Simonyan, A. Zisserman, Very deep convolutional networks for large-scale image recognition, preprint, arXiv: 1409.1556. |
[29] | M. Tan, Q. Le, Efficientnet: Rethinking model scaling for convolutional neural networks, preprint, arXiv: 1905.11946. |
[30] | G. Huang, Z. Liu, L. Van Der Maaten, K. Weinberger, Densely connected convolutional networks, in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, (2017), 2261–2269. https://doi.org/10.1109/CVPR.2017.243 |
[31] | Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, et al., Swin transformer: Hierarchical vision transformer using shifted windows, in 2021 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, (2021), 9992–10002. https://doi.org/10.1109/ICCV48922.2021.00986 |
1. | Xin He, Xin Zhao, Tao Feng, Zhipeng Qiu, Dynamical behaviors of a prey-predator model with foraging arena scheme in polluted environments, 2021, 71, 1337-2211, 235, 10.1515/ms-2017-0463 | |
2. | R. Rajaji, M. Pitchaimani, Analysis of stochastic viral infection model with lytic and nonlytic immune responses, 2020, 38, 0736-2994, 490, 10.1080/07362994.2019.1702559 | |
3. | Fathalla A. Rihan, Hebatallah J. Alsakaji, Persistence and extinction for stochastic delay differential model of prey predator system with hunting cooperation in predators, 2020, 2020, 1687-1847, 10.1186/s13662-020-02579-z | |
4. | Yongzhong Lan, Jianping Shi, Hui Fang, Hopf Bifurcation and Control of a Fractional-Order Delay Stage Structure Prey-Predator Model with Two Fear Effects and Prey Refuge, 2022, 14, 2073-8994, 1408, 10.3390/sym14071408 | |
5. | Feng Rao, Yun Kang, Dynamics of a stochastic prey–predator system with prey refuge, predation fear and its carry-over effects, 2023, 175, 09600779, 113935, 10.1016/j.chaos.2023.113935 | |
6. | Subarna Roy, Subhas Khajanchi, Pankaj Kumar Tiwari, Fear and its carry-over effects in a generalist predator–prey system featuring cooperative hunting, 2025, 1598-5865, 10.1007/s12190-024-02346-3 |
Equilibrium | Existence Condition | Stability Condition |
Always exists | Always saddle | |
Always exists | Sink if Saddle if | |
Always sink |