Agricultural decision-making involves a complex process of choosing strategies and options to enhance resource utilization, overall productivity, and farming practices. Agricultural stakeholders and farmers regularly make decisions at various levels of the farm cycle, ranging from crop selection and planting to harvesting and marketing. In agriculture, where crop health has played a central role in economic and yield outcomes, incorporating deep learning (DL) techniques has developed as a transformative force for the decision-making process. DL techniques, with their capability to discern subtle variations and complex patterns in plant health, empower agricultural experts and farmers to make informed decisions based on data-driven, real-time insights. Thus, we presented a Bayesian optimizer with deep learning based pepper leaf disease detection for decision making (BODL-PLDDM) approach in the agricultural sector. The BODL-PLDDM technique aimed to identify the healthy and bacterial spot pepper leaf disease. Primarily, the BODL-PLDDM technique involved a Wiener filtering (WF) approach for pre-processing. Besides, the complex and intrinsic feature patterns could be extracted using the Inception v3 model. Also, the Bayesian optimization (BO) algorithm was used for the optimal hyperparameter selection process. For disease detection, a crayfish optimization algorithm (COA) with a long short-term memory (LSTM) method was employed to identify the precise presence of pepper leaf diseases. The experimentation validation of the BODL-PLDDM system was verified using the Plant Village dataset. The obtained outcomes underlined the promising detection results of the BODL-PLDDM technique over other existing methods.
Citation: Asma A Alhashmi, Manal Abdullah Alohali, Nazir Ahmad Ijaz, Alaa O. Khadidos, Omar Alghushairy, Ahmed Sayed. Bayesian optimization with deep learning based pepper leaf disease detection for decision-making in the agricultural sector[J]. AIMS Mathematics, 2024, 9(7): 16826-16847. doi: 10.3934/math.2024816
[1] | Abdelkrim Salim, Sabri T. M. Thabet, Imed Kedim, Miguel Vivas-Cortez . On the nonlocal hybrid (k,φ)-Hilfer inverse problem with delay and anticipation. AIMS Mathematics, 2024, 9(8): 22859-22882. doi: 10.3934/math.20241112 |
[2] | Kangqun Zhang . Existence and uniqueness of positive solution of a nonlinear differential equation with higher order Erdélyi-Kober operators. AIMS Mathematics, 2024, 9(1): 1358-1372. doi: 10.3934/math.2024067 |
[3] | Dumitru Baleanu, Babak Shiri . Nonlinear higher order fractional terminal value problems. AIMS Mathematics, 2022, 7(5): 7489-7506. doi: 10.3934/math.2022420 |
[4] | Yujun Cui, Chunyu Liang, Yumei Zou . Existence and uniqueness of solutions for a class of fractional differential equation with lower-order derivative dependence. AIMS Mathematics, 2025, 10(2): 3797-3818. doi: 10.3934/math.2025176 |
[5] | Dumitru Baleanu, Babak Shiri . Generalized fractional differential equations for past dynamic. AIMS Mathematics, 2022, 7(8): 14394-14418. doi: 10.3934/math.2022793 |
[6] | Qi Wang, Chenxi Xie, Qianqian Deng, Yuting Hu . Controllability results of neutral Caputo fractional functional differential equations. AIMS Mathematics, 2023, 8(12): 30353-30373. doi: 10.3934/math.20231550 |
[7] | Wedad Albalawi, Muhammad Imran Liaqat, Fahim Ud Din, Kottakkaran Sooppy Nisar, Abdel-Haleem Abdel-Aty . Significant results in the pth moment for Hilfer fractional stochastic delay differential equations. AIMS Mathematics, 2025, 10(4): 9852-9881. doi: 10.3934/math.2025451 |
[8] | Jun Moon . The Pontryagin type maximum principle for Caputo fractional optimal control problems with terminal and running state constraints. AIMS Mathematics, 2025, 10(1): 884-920. doi: 10.3934/math.2025042 |
[9] | Wei Zhang, Jifeng Zhang, Jinbo Ni . New Lyapunov-type inequalities for fractional multi-point boundary value problems involving Hilfer-Katugampola fractional derivative. AIMS Mathematics, 2022, 7(1): 1074-1094. doi: 10.3934/math.2022064 |
[10] | Dehong Ji, Weigao Ge . A nonlocal boundary value problems for hybrid ϕ-Caputo fractional integro-differential equations. AIMS Mathematics, 2020, 5(6): 7175-7190. doi: 10.3934/math.2020459 |
Agricultural decision-making involves a complex process of choosing strategies and options to enhance resource utilization, overall productivity, and farming practices. Agricultural stakeholders and farmers regularly make decisions at various levels of the farm cycle, ranging from crop selection and planting to harvesting and marketing. In agriculture, where crop health has played a central role in economic and yield outcomes, incorporating deep learning (DL) techniques has developed as a transformative force for the decision-making process. DL techniques, with their capability to discern subtle variations and complex patterns in plant health, empower agricultural experts and farmers to make informed decisions based on data-driven, real-time insights. Thus, we presented a Bayesian optimizer with deep learning based pepper leaf disease detection for decision making (BODL-PLDDM) approach in the agricultural sector. The BODL-PLDDM technique aimed to identify the healthy and bacterial spot pepper leaf disease. Primarily, the BODL-PLDDM technique involved a Wiener filtering (WF) approach for pre-processing. Besides, the complex and intrinsic feature patterns could be extracted using the Inception v3 model. Also, the Bayesian optimization (BO) algorithm was used for the optimal hyperparameter selection process. For disease detection, a crayfish optimization algorithm (COA) with a long short-term memory (LSTM) method was employed to identify the precise presence of pepper leaf diseases. The experimentation validation of the BODL-PLDDM system was verified using the Plant Village dataset. The obtained outcomes underlined the promising detection results of the BODL-PLDDM technique over other existing methods.
In 1998, Kahlig and Matkowski [1] proved in particular that every homogeneous bivariable mean M in (0,∞) can be represented in the form
M(x,y)=A(x,y)fM,A(x−yx+y), |
where A is the arithmetic mean and fM,A: (−1,1)⟶(0,2) is a unique single variable function (with the graph laying in a set of a butterfly shape), called an A-index of M.
In this paper we consider Seiffert function f:(0,1)→R which fulfils the following condition
t1+t≤f(t)≤t1−t. |
According to the results of Witkowski [2] we introduce the mean Mf of the form
Mf(x,y)={|x−y|2f(|x−y|x+y)x≠y,xx=y. | (1.1) |
In this paper a mean Mf:R2+⟶R is the function that is symmetric, positively homogeneous and internal in sense [2]. Basic result of Witkowski is correspondence between a mean Mf and Seiffert function f of Mf is given by the following formula
f(t)=tMf(1−t,1+t), | (1.2) |
where
t=|x−y|x+y. | (1.3) |
Therefore, f and Mf form a one-to-one correspondence via (1.1) and (1.2). For this reason, in the following we can rewrite f=:fM.
Throughout this article, we say x≠y, that is, t∈(0,1). For convenience, we note that M1<M2 means M1(x,y)<M2(x,y) holds for two means M1 and M2 with x≠y. Then there is a fact that the inequality fM1(t)>fM2(t) holds if and only if M1<M2. That is to say,
1fM1<1fM2⟺M1<M2. | (1.4) |
The above relationship (1.4) inspires us to ask a question: Can we transform the means inequality problem into the reciprocal inequality problem of the corresponding Seiffert functions? Witkowski [2] answers this question from the perspective of one-to-one correspondence. We find that these two kinds of inequalities are equivalent in similar linear inequalities. We describe this result in Lemma 2.1 as a support of this paper.
As we know, the study of inequalities for mean values has always been a hot topic in the field of inequalities. For example, two common means can be used to define some new means. The recent success in this respect can be seen in references [3,4,5,6,7,8]. In [2], Witkowski introduced the following two new means, one called sine mean
Msin(x,y)={|x−y|2sin(|x−y|x+y)x≠yxx=y, | (1.5) |
and the other called hyperbolic tangent mean
Mtanh(x,y)={|x−y|2tanh(|x−y|x+y)x≠yxx=y. | (1.6) |
Recently, Nowicka and Witkowski [9] determined various optimal bounds for the Msin(x,y) and Mtanh(x,y) by the arithmetic mean A(x,y)=(x+y)/2 and centroidal mean
Ce(x,y)=23x2+xy+y2x+y |
as follows:
Proposition 1.1. The double inequality
(1−α)A+αCe<Msin<(1−β)A+βCe |
holds if and only if α≤ 1/2 and β≥(3/sin1)−3≈0.5652.
Proposition 1.2. The double inequality
(1−α)A+αCe<Mtanh<(1−β)A+βCe |
holds if and only if α≤ (3/tanh1)−3≈0.9391 and β≥1.
Proposition 1.3. The double inequality
(1−α)C−1e+αA−1<M−1sin<(1−β)C−1e+βA−1 |
holds if and only if α≤ 4sin1−3≈0.3659 and β≥1/2.
Proposition 1.4. The double inequality
(1−α)C−1e+αA−1<M−1tanh<(1−β)C−1e+βA−1 |
holds if and only if α≤ 0 and β≥4tanh1−3≈0.0464.
Proposition 1.5. The double inequality
(1−α)A2+αC2e<M2sin<(1−β)A2+βC2e |
holds if and only if α≤ 1/2 and β≥(9cot21)/7≈0.5301.
Proposition 1.6. The double inequality
(1−α)A2+αC2e<M2tanh<(1−β)A2+βC2e |
holds if and only if α≤ (9(coth21−1))/7≈0.9309 and β≥1.
Proposition 1.7. The double inequality
(1−α)C−2e+αA−2<M−2sin<(1−β)C−2e+βA−2 |
holds if and only if α≤(16sin21−9)/7≈0.3327 and β≥1/2.
Proposition 1.8. The double inequality
(1−α)C−2e+αA−2<M−2tanh<(1−β)C−2e+βA−2 |
holds if and only if α≤0 and β≥(16tanh21−9)/7≈0.0401.
In essence, the above results are how the two new means Msin and Mtanh are expressed linearly, harmoniously, squarely, and harmoniously in square by the two classical means Ce(x,y) and A(x,y). In this paper, we study the following two-sided inequalities in exponential form for nonzero number p∈R
(1−αp)Ap+αpCep<Mpsin<(1−βp)Ap+βpCep, | (1.7) |
(1−λp)Ap+λpCep<Mptanh<(1−μp)Ap+μpCep | (1.8) |
in order to reach a broader conclusion including all the above properties. The main conclusions of this paper are as follows:
Theorem 1.1. Let x,y>0, x≠y, p≠0 and
p♣=3cos2+sin2+13sin2−cos2−3≈4.588. |
Then the following are considered.
(i) If p≥p♣, the double inequality
(1−αp)Ap+αpCep<Mpsin<(1−βp)Ap+βpCep | (1.9) |
holds if and only if αp≤3p(1−sinp1)/[(sinp1)(4p−3p)] and βp≥1/2.
(ii) If 0<p≤12/5, the double inequality
(1−αp)Ap+αpCep<Mpsin<(1−βp)Ap+βpCep | (1.10) |
holds if and only if αp≤1/2 and βp≥3p(1−sinp1)/[(sinp1)(4p−3p)].
(iii) If p<0, the double inequality
(1−βp)Ap+βpCep<Mpsin<(1−αp)Ap+αpCep | (1.11) |
holds if and only if αp≤1/2 and βp≥3p(1−sinp1)/[(sinp1)(4p−3p)].
Theorem 1.2. Let x,y>0, x≠y, p≠0 and
p∗=−16cosh2−3cosh4+4sinh2+3cosh4−12sinh2+15≈−3.4776. |
Then the following are considered:
(i) If p>0, the double inequality
(1−λp)Ap+λpCep<Mptanh<(1−μp)Ap+μpCep | (1.12) |
holds if and only if λp≤((coth1)p−1)/((4/3)p−1) and μp≥1.
(ii) If p∗≤p<0,
(1−μp)Ap+μpλpCep<Mptanh<(1−λp)Ap+λpCep | (1.13) |
holds if and only if λp≤((coth1)p−1)/((4/3)p−1) and μp≥1.
We first introduce a theoretical support of this paper.
Lemma 2.1. ([10]) Let K(x,y),R(x,y), and N(x,y) be three means with two positive distinct parameters x and y; fK(t), fR(t), and fN(t) be the corresponding Seiffert functions of the former, ϑ1,ϑ2,θ1,θ2,p∈R, and p≠0. Then
ϑ1Kp(x,y)+ϑ2Np(x,y)≤Rp(x,y)≤θ1Kp(x,y)+θ2Np(x,y) | (2.1) |
⟺ϑ1fpK(t)+ϑ2fpN(t)≤1fpR(t)≤θ1fpK(t)+θ2fpN(t). | (2.2) |
It must be mentioned that the key steps to prove the above results are following:
Mf(u,v)=Mf(λ2xx+y,λ2yx+y)=λMf(2xx+y,2yx+y)=λMf(1−t,1+t)=λtfM(t), | (2.3) |
where
{u=λ2xx+yv=λ2yx+y, 0<x<y, λ>0. |
and 0<t<1,
t=y−xx+y. |
In order to prove the main conclusions, we shall introduce some very suitable methods which are called the monotone form of L'Hospital's rule (see Lemma 2.2) and the criterion for the monotonicity of the quotient of power series (see Lemma 2.3).
Lemma 2.2. ([11,12]) For −∞<a<b<∞, let f,g:[a,b]→R be continuous functions that are differentiable on (a,b), with f(a)=g(a)=0 or f(b)=g(b)=0. Assume that g′(t)≠0 for each x in (a,b). If f′/g′ is increasing (decreasing) on (a,b), then so is f/g.
Lemma 2.3. ([13]) Let an and bn (n=0,1,2,⋅⋅⋅) be real numbers, and let the power series A(x)=∑∞n=0anxn and B(x)=∑∞n=0bnxn be convergent for |x|<R (R≤+∞). If bn>0 for n=0,1,2,⋅⋅⋅, and if εn=an/bn is strictly increasing (or decreasing) for n=0,1,2,⋅⋅⋅, then the function A(x)/B(x) is strictly increasing (or decreasing) on (0,R) (R≤+∞).
Lemma 2.4. ([14,15]) Let B2n be the even-indexed Bernoulli numbers. Then we have the following power series expansions
cotx=1x−∞∑n=122n(2n)!|B2n|x2n−1, 0<|x|<π, | (2.4) |
1sin2x=csc2x=−(cotx)′=1x2+∞∑n=122n(2n−1)(2n)!|B2n|x2n−2, 0<|x|<π. | (2.5) |
Lemma 2.5. ([16,17,18,19,20]) Let B2n the even-indexed Bernoulli numbers, n=1,2,…. Then
22n−1−122n+1−1(2n+2)(2n+1)π2<|B2n+2||B2n|<22n−122n+2−1(2n+2)(2n+1)π2. |
Lemma 2.6. Let l1(t) be defined by
l1(t)=s1(t)r1(t), |
where
s1(t)=6t2+2t4−12sin2t−2t3costsint+6tcostsintsin2t,r1(t)=8t2sin2t+2t4sin2t−6t2−2t4−6sin2t+12tcostsintsin2t. |
Then the double inequality
125<l1(t)<p♣=3cos2+sin2+13sin2−cos2−3≈4.588 | (2.6) |
holds for all t∈(0,1), where the constants 12/5 and (3cos2+sin2+1)/(3sin2−cos2−3)≈4.588 are the best possible in (2.6).
Proof. Since
1l1(t)=r1(t)s1(t), |
and
r1(t)=8t2sin2t+2t4sin2t−6t2−2t4−6sin2t+12tcostsintsin2t=8t2−2t41sin2t−6t21sin2t+2t4+12tcostsint−6=8t2−2t4[1t2+∞∑n=122n(2n−1)(2n)!|B2n|t2n−2]−6t2[1t2+∞∑n=122n(2n−1)(2n)!|B2n|t2n−2]+2t4+12t[1t−∞∑n=122n(2n)!|B2n|t2n−1]−6=23t4−∞∑n=3[22n−1(2n−3)(2n−2)!|B2n−2|+6⋅22n(2n+1)(2n)!|B2n|]t2n=:∞∑n=2ant2n, |
where
a2=23,an=−[22n−1(2n−3)(2n−2)!|B2n−2|+6⋅22n(2n+1)(2n)!|B2n|], n=3,4,…, |
s1(t)=6t2+2t4−12sin2t−2t3costsint+6tcostsintsin2t=6t21sin2t+2t41sin2t+6tcostsint−2t3costsint−12=6t2[1t2+∞∑n=122n(2n−1)(2n)!|B2n|t2n−2]+2t4[1t2+∞∑n=122n(2n−1)(2n)!|B2n|t2n−2]+6t[1t−∞∑n=122n(2n)!|B2n|t2n−1]−2t3[1t−∞∑n=122n(2n)!|B2n|t2n−1]−12=∞∑n=212⋅22n(n−1)(2n)!|B2n|t2n+∞∑n=14n⋅22n(2n)!|B2n|t2n+2=∞∑n=212⋅22n(n−1)(2n)!|B2n|t2n+∞∑n=2(n−1)⋅22n(2n−2)!|B2n−2|t2n=∞∑n=2[12⋅22n(n−1)(2n)!|B2n|+(n−1)⋅22n(2n−2)!|B2n−2|]t2n=85t4+∞∑n=3[12⋅22n(n−1)(2n)!|B2n|+(n−1)⋅22n(2n−2)!|B2n−2|]t2n=:∞∑n=2bnt2n, |
where
b2=85,bn=12⋅22n(n−1)(2n)!|B2n|+(n−1)⋅22n(2n−2)!|B2n−2|>0, n=3,4,…. |
Setting
qn=anbn, n=2,3,…, |
we have
q2=512=0.41667,qn=−22n−1(2n−3)(2n−2)!|B2n−2|+6⋅22n(2n+1)(2n)!|B2n|12⋅22n(n−1)(2n)!|B2n|+(n−1)⋅22n(2n−2)!|B2n−2|, n=3,4,…. |
Here we prove that the sequence {qn}n≥2 decreases monotonously. Obviously, q2>0>q3. We shall prove that for n≥3,
qn>qn+1⟺−22n−1(2n−3)(2n−2)!|B2n−2|+6⋅22n(2n+1)(2n)!|B2n|12⋅22n(n−1)(2n)!|B2n|+(n−1)⋅22n(2n−2)!|B2n−2|>−22n+1(2n−1)(2n)!|B2n|+6⋅22n+2(2n+3)(2n+2)!|B2n+2|12⋅22n+2n(2n+2)!|B2n+2|+n⋅22n+2(2n)!|B2n|⟺22n−1(2n−3)(2n−2)!|B2n−2|+6⋅22n(2n+1)(2n)!|B2n|12⋅22n(n−1)(2n)!|B2n|+(n−1)⋅22n(2n−2)!|B2n−2|<22n+1(2n−1)(2n)!|B2n|+6⋅22n+2(2n+3)(2n+2)!|B2n+2|12⋅22n+2n(2n+2)!|B2n+2|+n⋅22n+2(2n)!|B2n|, |
that is,
2(2n)!(2n−2)!|B2n−2||B2n|+24(4n−3)(2n−2)!(2n+2)!|B2n+2||B2n−2||B2n||B2n|>24(4n−1)((2n)!)2+864(2n)!(2n+2)!|B2n+2||B2n|. | (2.7) |
By Lemma 2.5 we have
2(2n)!(2n−2)!|B2n−2||B2n|+24(4n−3)(2n−2)!(2n+2)!|B2n+2||B2n−2||B2n||B2n|>2(2n)!(2n−2)!22n−122n−2−1π2(2n)(2n−1)+24(4n−3)(2n−2)!(2n+2)!22n−1−122n+1−1(2n+2)(2n+1)π222n−122n−2−1π2(2n)(2n−1)=2π2(2n)!(2n)!22n−122n−2−1+24(4n−3)(2n)!(2n)!22n−1−122n+1−122n−122n−2−1, |
and
24(4n−1)(2n)!2+864(2n)!(2n+2)!|B2n+2||B2n|<24(4n−1)(2n)!2+864(2n)!(2n+2)!22n−122n+2−1(2n+2)(2n+1)π2=24(4n−1)(2n)!2+864(2n)!(2n)!22n−122n+2−11π2. |
So we can complete the prove (2.7) when proving
2π2(2n)!(2n)!22n−122n−2−1+24(4n−3)(2n)!(2n)!22n−1−122n+1−122n−122n−2−1>24(4n−1)((2n)!)2+864(2n)!(2n)!22n−122n+2−11π2 |
or
2π2(22n−1)22n−2−1+24(4n−3)22n−1−122n+1−122n−122n−2−1>24(4n−1)+22n−122n+2−1864π2. |
In fact,
2π2(22n−1)22n−2−1+24(4n−3)22n−1−122n+1−122n−122n−2−1−[24(4n−1)+22n−122n+2−1864π2]=:8H(n)π2(22n+2−1)(22n−4)(22n+1−1), |
where
H(n)=8⋅26n(π+3)(π−3)(π2+3)+2⋅24n(72π2n+60π2−7π4+594)−22n(36π2n+123π2−7π4+1404)+(24π2−π4+432)>0 |
for all n≥3.
So the sequence {qn}n≥2 decreases monotonously. By Lemma 2.3 we obtain that r1(t)/s1(t) is decreasing on (0,1), which means that the function l1(t) is increasing on (0,1). In view of
limt→0+l1(t)=125 and limt→1−l1(t)=p♣=3cos2+sin2+13sin2−cos2−3≈4.588, |
the proof of this lemma is complete.
Lemma 2.7. Let l2(t) be defined by
l2(t)=2⋅3cosh4t−12t2cosh2t−4t4cosh2t+2t3sinh2t−6tsinh2t−3t2cosh4t−3cosh4t+24tsinh2t−25t2−8t4+3=:2B(t)A(t), 0<t<∞, |
where
A(t)=t2cosh4t−3cosh4t+24tsinh2t−25t2−8t4+3,B(t)=3cosh4t−12t2cosh2t−4t4cosh2t+2t3sinh2t−6tsinh2t−3. |
Then l2(t) is strictly decreasing on (0,∞).
Proof. Let's take the power series expansions
sinhkt=∞∑n=0k2n+1(2n+1)!t2n+1, coshkt=∞∑n=0k2n(2n)!t2n |
into A(t) and B(t), and get
A(t)=∞∑n=2cnt2n+2, B(t)=∞∑n=2dnt2n+2, |
where
c2=0,cn=[2(3n+2n2−23)22n+48(2n+2)(2n+2)!]22n, n=3,4,…,dn=[48⋅22n−8(n+1)(5n−n2+2n3+6)(2n+2)!]22n, n=2,3,…, |
Setting
kn=cndn=48(n+1)+22n(3n+2n2−23)4(6⋅22n−11n−4n2−n3−2n4−6), n=2,3,…, |
Here we prove that the sequence {kn}n≥2 decreases monotonously. Obviously, k2=0<k3. For n≥3,
kn<kn+1⟺48(n+1)+22n(3n+2n2−23)4(6⋅22n−11n−4n2−n3−2n4−6)<48(n+2)+22n+2(3(n+1)+2(n+1)2−23)4(6⋅22n+2−11(n+1)−4(n+1)2−(n+1)3−2(n+1)4−6)⟺48(n+1)+22n(3n+2n2−23)6⋅22n−11n−4n2−n3−2n4−6<48n+96+22n+2(7n+2n2−18)6⋅22n+2−30n−19n2−9n3−2n4−24 |
follows from Δ(n)>0 for all n≥2, where
Δ(n)=(48n+96+22n+2(7n+2n2−18))(6⋅22n−11n−4n2−n3−2n4−6)−(48(n+1)+22n(3n+2n2−23))(6⋅22n+2−30n−19n2−9n3−2n4−24)=24⋅24n(4n+5)−22n(858n+367n2+218n3−103n4+40n5+12n6+696)+1248n+1440n2+1056n3+288n4+576=:22n[j(n)22n−i(n)]+w(n) |
with
j(n)=24(4n+5),i(n)=858n+367n2+218n3−103n4+40n5+12n6+696,w(n)=1248n+1440n2+1056n3+288n4+576>0. |
We have that Δ(2)=5376>0 and shall prove that
j(n)22n−i(n)>0⟺22n>i(n)j(n) | (2.8) |
holds for all n≥3. Now we use mathematical induction to prove (2.8). When n=3, the left-hand side and right-hand side of (2.8) are 26=64 and i(3)/j(3)=941/17≈55.353, which implies (2.8) holds for n=3. Assuming that (2.8) holds for n=m, that is,
22m>i(m)j(m). | (2.9) |
Next, we prove that (2.8) is valid for n=m+1. By (2.9) we have
22(m+1)=4⋅22m>4i(m)j(m), |
in order to complete the proof of (2.8) it suffices to show that
4i(m)j(m)>i(m+1)j(m+1)⟺4i(m)j(m+1)−i(m+1)j(m)>0. |
In fact,
4i(m)j(m+1)−i(m+1)j(m)=17280m7+90720m6−60000m5−97176m4+1169232m3+2266104m2+3581136m+2154816=146337408+234401616(m−3)+189746328(m−3)2+92580720(m−3)3+27579624(m−3)4+4838880(m−3)5+453600(m−3)6+17280(m−3)7>0 |
for m≥3 due to the coefficients of the power square of (m−1) are positive.
By Lemma 2.3 we get that A(t)/B(t) is strictly increasing on (0,∞). So the function l2(x) is strictly decreasing on (0,∞).
The proof of Lemma 2.7 is complete.
Via (1.3) and (1.2) we can obtain
fA(t)=t,fCe(t)=3t3+t2,fMsin(t)=sint,fMtanh(t)=tanht. |
Then by Lemma 2.1 and (2.3) we have
αp<Mpsin−ApCep−Ap<βp⟺αp<(1sint)p−(1t)p(3+t23t)p−(1t)p<βp,λp<Mptanh−ApCep−Ap<μp⟺λp<(1tanht)p−(1t)p(3+t23t)p−(1t)p<μp. |
So we turn to the proof of the following two theorems.
Theorem 3.1. Let t∈(0,1) and
p♣=3cos2+sin2+13sin2−cos2−3≈4.588. |
Then,
(i) if p≥p♣, the double inequality
αp<(1sint)p−(1t)p(3+t23t)p−(1t)p<βp | (3.1) |
holds if and only if αp≤3p(1−sinp1)/[(sinp1)(4p−3p)] and βp≥1/2;
(ii) if 0≠p≤12/5=2.4 and p≠0 the double inequality
βp<(1sint)p−(1t)p(3+t23t)p−(1t)p<αp | (3.2) |
holds if and only if αp≤1/2 and β≥3p(1−sinp1)/[(sinp1)(4p−3p)].
Theorem 3.2. Let t∈(0,1) and
p∗=−16cosh2−3cosh4+4sinh2+3cosh4−12sinh2+15≈−3.4776. |
If 0≠p≥−3.4776, the double inequality
λp<(1tanht)p−(1t)p(3+t23t)p−(1t)p<μp | (3.3) |
holds if and only if λp≤((coth1)p−1)/((4/3)p−1) and μp≥1.
Let
F(t)=(1sint)p−(1t)p(3+t23t)p−(1t)p=(tsint)p−1(3+t23)p−1=:f(t)g(t)=f(t)−f(0+)g(t)−g(0+), |
where
f(t)=(tsint)p−1,g(t)=(3+t23)p−1. |
Then
f′(t)=psin2t(sint−tcost)(tsint)p−1,g′(t)=23(13)p−1pt(t2+3)p−1, |
f′(t)g′(t)=3p21tsin2t(sint−tcost)(t(t2+3)sint)p−1, |
and
(f′(t)g′(t))′=14(3t(sint)(t2+3))pr1(t)t3sin2t[s1(t)r1(t)−p]=:14(3t(sint)(t2+3))pr1(t)t3sin2t[l1(t)−p], |
where the three functions s1(t), r1(t), and l1(t) are shown in Lemma 2.6.
By Lemma 2.6 we can obtain the following results:
(a) When p≥maxt∈(0,1)l1(t)=:p♣=(3cos2+sin2+1)/(3sin2−cos2−3)≈4.588,
(f′(t)g′(t))′≤0⟹f′(t)g′(t) is decreasing on (0,1), |
this leads to F(t)=f(t)/g(t) is decreasing on (0,1) by Lemma 2.1. In view of
F(0+)=12, F(1−)=3p(1−sinp1)(sinp1)(4p−3p), | (3.4) |
we have that (3.1) holds.
(b) When 0≠p≤12/5=mint∈(0,1)l1(t),
(f′(t)g′(t))′≥0⟹f′(t)g′(t) is increasing on (0,1), |
this leads to F(t)=f(t)/g(t) is increasing on (0,1) by Lemma 2.2. In view of (3.4) we have that (3.2) holds.
The proof of Theorem 3.1 is complete.
Let
G(t)=(1tanht)p−(1t)p(3+t23t)p−(1t)p=(ttanht)p−1(3+t23)p−1=:u(t)v(t)=u(t)−u(0+)v(t)−v(0+). |
Then
u′(t)=ptanh2t(ttanht)p−1(ttanh2t+tanht−t),v′(t)=23pt(t2+33)p−1, |
u′(t)v′(t)=32ttanh2t+tanht−tttanh2t[3t(t2+3)tanht]p−1, |
and
(u′(t)v′(t))′=−116[3tcosht(3+t2)sinht]pA(t)t3cosh2tsinh2t[p+2B(t)A(t)]=:−116[3tcosht(3+t2)sinht]pA(t)t3cosh2tsinh2t[p+l2(t)], |
where the three functions A(t), B(t), and l2(t) are shown in Lemma 2.7. By Lemma 2.7 we see that l2(x) is strictly decreasing on (0,1). Since
limt→0+l2(t)=∞,limt→1−l2(t)=16cosh2−3cosh4+4sinh2+3cosh4−12sinh2+15=:p#≈3.4776, |
we obtain the following result:
When p≥maxt∈(0,1){−l2(t)}=−p#=:p∗≈−3.4776,
(u′(t)v′(t))′≤0⟹u′(t)v′(t) is decreasing on (0,1), |
this leads to G(t)=u(t)/v(t) is decreasing on (0,1) by Lemma 2.2. Since
G(0+)=1,G(1−)=(cosh1sinh1)p−1(43)p−1, |
we have
G(1−)<G(t)<G(0+), |
which completes the proof of Theorem 3.2.
Remark 4.1. Letting p=1,−1,2,−2 in Theorems 1.1 and 1.2 respectively, one can obtain Propositions 1.1–1.8.
From Theorems 1.1 and 1.2, we can also get the following important conclusions:
Corollary 4.1. Let x,y>0, x≠y, and
p♣=3cos2+sin2+13sin2−cos2−3≈4.588,α=3p♣(1−sinp♣1)(sinp♣1)(4p♣−3p♣)≈0.44025,β=12. |
Then the double inequality
(1−α)Ap♣+αCep♣<Mp♣sin<(1−β)Ap♣+βCep♣ | (4.1) |
holds, where the constants α and β are the best possible in (4.1).
Corollary 4.2. Let x,y>0, x≠y, and
θ=12,ϑ=312/5(1−sin12/51)(sin12/51)(412/5−312/5)≈0.51603. |
Then the double inequality
(1−θ)A12/5+θCe12/5<M12/5sin<(1−ϑ)A12/5+ϑCe12/5 | (4.2) |
holds, where the constants θ and ϑ are the best possible in (4.2).
Corollary 4.3. Let x,y>0, x≠y, and
p∗=−16cosh2−3cosh4+4sinh2+3cosh4−12sinh2+15≈−3.4776,λ=(coth1)p∗−1(4/3)p∗−1≈0.96813,μ=1. |
Then the double inequality
(1−μ)Ap∗+μCep∗<Mp∗tanh<(1−λ)Ap∗+λCep∗ | (4.3) |
holds, where the constants λ and μ are the best possible in (4.3).
In this paper, we have studied exponential type inequalities for Msin and Mtanh in term of A and Ce for nonzero number p∈R:
(1−αp)Ap+αpCep<Mpsin<(1−βp)Ap+βpCep,(1−λp)Ap+λpCep<Mptanh<(1−μp)Ap+μpCep, |
obtained a lot of interesting conclusions which include the ones of the previous similar literature. In fact, we can consider similar inequalities for dual means of the two means Msin and Mtanh, and we can replace A and Ce by other famous means. Therefore, the content of this research is very extensive.
The authors are grateful to editor and anonymous referees for their careful corrections to and valuable comments on the original version of this paper.
The first author was supported by the National Natural Science Foundation of China (no. 61772025). The second author was supported in part by the Serbian Ministry of Education, Science and Technological Development, under projects ON 174032 and III 44006.
The authors declare that they have no conflict of interest.
[1] | R. Sharma, K. Vinay, D. Bordoloi, Deep learning meets agriculture: A faster RCNN based approach to pepper leaf blight disease detection and multi-classification, In: 2023 4th International Conference for Emerging Technology (INCET), 2023. |
[2] | M. B. Devi, K. Amarendra, Machine learning-based application to detect pepper leaf diseases using histgradientboosting classifier with fused HOG and LBP features, In: Smart Technologies in Data Science and Communication: Proceedings of SMART-DSC, Singapore: Springer, 2021. |
[3] | T. H. H. Aldhyani, A. Hasan, R. J. Eunice, D. J. Hemanth, Leaf pathology detection in potato and pepper bell plant using convolutional neural networks, In: 2022 7th International Conference on Communication and Electronics Systems (ICCES), 2022. |
[4] | C. H. Kim, M. N. R. Samsuzzaman, K. Y. Lee, M. R. Ali, Deep learning-based identification of Pepper (Capsicum annuum L.) diseases: A review, Precis. Agric., 5 (2023), 68. |
[5] | A. S. Kini, K. V. Prema, S. N. Pai, State of the art deep learning implementation for multiclass classification of black pepper leaf diseases, 2023. https://doi.org/10.21203/rs.3.rs-3272019/v1 |
[6] | I. Haque, M. A. Islam, K. Roy, M. M. Rahaman, A. A. Shohan, I. Md Saiful, Classifying pepper disease based on transfer learning: A deep learning approach, In: 2022 International Conference on Applied Artificial Intelligence and Computing (ICAAIC), 2022. http://doi.org/10.1109/ICAAIC53929.2022.9793178 |
[7] | K. Andersson, M. S. Hoassain, Bell pepper leaf disease classification using convolutional neural network, In: Intelligent Computing & Optimization: Proceedings of the 5th International Conference on Intelligent Computing and Optimization 2022 (ICO2022), Springer Nature, 569 (2022). |
[8] | Y. Akhalifi, A. Subekti, Bell pepper leaf disease classification using fine-tuned transfer learning, J. Elektronikadan Telekomunikasi, 23 (2023), 55–61. |
[9] | P. Thakur, C. Anuradha, A. P. Singh, Plant disease detection of bell pepper plant using transfer learning over different models, In: 2021 8th International Conference on Signal Processing and Integrated Networks (SPIN), 2021,384–389. |
[10] | C. Y. Khew, Y. Q. Teow, E. T. Lau, S. S. Hwang, C. H. Bong, N. K. Lee, Evaluation of deep learning for image-based black pepper disease and nutrient deficiency classification, In: 2021 2nd International Conference on Artificial Intelligence and Data Sciences (AiDAS), 2021, 1–6. http://doi.org/10.1109/AiDAS53897.2021.9574346 |
[11] | I. Bouacida, B. Farou, L. Djakhdjakha, H. Seridi, M. Kurulay, Innovative deep learning approach for cross-crop plant disease detection: A generalized method for identifying unhealthy leaves, Inform. Process. Agricul., 2024. |
[12] | M. Shoaib, T. Hussain, B. Shah, I. Ullah, S. M. Shah, F. Ali, et al., Deep learning-based segmentation and classification of leaf images for detection of tomato plant disease, Front. Plant Sci., 13 (2022), 1031748. |
[13] | Y. A. Bezabih, A. O. Salau, B. M. Abuhayi, A. A. Mussa, A. M. Ayalew, CPD-CCNN: Classification of pepper disease using a concatenation of convolutional neural network models, Sci. Rep., 13 (2023), 15581. |
[14] | S. S. A. Begum, H. Syed, GSAtt-CMNetV3: Pepper leaf disease classification using osprey optimization, IEEE Access, 2024. |
[15] | D. I. Lee, J. H. Lee, S. H. Jang, S. J. Oh, I. C. Doo, Crop disease diagnosis with deep learning-based image captioning and object detection, Appl. Sci., 13 (2023), 3148. |
[16] | V. Gautam, R. K. Ranjan, P. Dahiya, A. Kumar, ESDNN: A novel ensembled stack deep neural network for mango leaf disease classification and detection, Multimed. Tools Appl., 83 (2024), 10989–11015. |
[17] | S. Ashwinkumar, S. Rajagopal, V. Manimaran, B. Jegajothi, Automated plant leaf disease detection and classification using optimal MobileNet based convolutional neural networks, Mater. Today Proceed., 51 (2022), 480–487. |
[18] | A. Kumar, V. K. Patel, Classification and identification of disease in potato leaf using hierarchical based deep learning convolutional neural network, Multimed. Tools Appl., 2023, 1–27. |
[19] | N. Krishnamoorthy, L. N. Prasad, C. P. Kumar, B. Subedi, H. B. Abraha, V. E. Sathishkumar, Rice leaf diseases prediction using deep neural networks with transfer learning, Environ. Res., 198 (2021), 111275. |
[20] |
M. Subramanian, N. P. Lv, S. VE, Hyperparameter optimization for transfer learning of VGG16 for disease identification in corn leaves using Bayesian optimization, Big Data, 10 (2022), 215–229. https://doi.org/10.1089/big.2021.021 doi: 10.1089/big.2021.021
![]() |
[21] |
J. V. Valls, D. Vivet, E. Chaumette, F. Vincent, P. Closas, Recursive linearly constrained Wiener filter for robust multi-channel signal processing, Signal Process., 167 (2020), 107291. https://doi.org/10.1016/j.sigpro.2019.107291 doi: 10.1016/j.sigpro.2019.107291
![]() |
[22] |
K. S. Rao, P. V. Terlapu, D. Jayaram, K. K. Raju, G. K. Kumar, R. Pemula, et al., Intelligent ultrasound imaging for enhanced breast cancer diagnosis: Ensemble transfer learning strategies, IEEE Access, 2024. http://doi.org/10.1109/ACCESS.2024.3358448 doi: 10.1109/ACCESS.2024.3358448
![]() |
[23] |
A. Riboni, N. Ghioldi, A. Candelieri, M. Borrotti, Bayesian optimization and deep learning for steering wheel angle prediction, Sci. Rep., 12 (2022), 8739. https://doi.org/10.1038/s41598-022-12509-6 doi: 10.1038/s41598-022-12509-6
![]() |
[24] |
A. Cuk, T. Bezdan, L. Jovanovic, M. Antonijevic, M. Stankovic, V. Simic, et al., Tuning attention based long-short term memory neural networks for Parkinson's disease detection using modified metaheuristics, Sci. Rep., 14 (2024), 4309. https://doi.org/10.1038/s41598-024-54680-y doi: 10.1038/s41598-024-54680-y
![]() |
[25] |
A. Sagheer, M. Kotb, Unsupervised pre-training of a deep LSTM-based stacked autoencoder for multivariate time series forecasting problems, Sci. Rep., 9 (2019), 19038. https://doi.org/10.1038/s41598-019-55320-6 doi: 10.1038/s41598-019-55320-6
![]() |
[26] | Plant Village dataset. Available from: https://www.kaggle.com/datasets/adilmubashirchaudhry/plant-village-dataset. |
1. | Miao-Kun Wang, Zai-Yin He, Tie-Hong Zhao, Qi Bao, Sharp weighted Hölder mean bounds for the complete elliptic integral of the second kind, 2022, 1065-2469, 1, 10.1080/10652469.2022.2155819 | |
2. | Ling Zhu, New Bounds for Arithmetic Mean by the Seiffert-like Means, 2022, 10, 2227-7390, 1789, 10.3390/math10111789 |