Page 192 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 6 Plant leaf disease classification based on feature selection 181

! ƒ! ƒ! ƒ!
X 3 ¼ X d A 3 $ D d (6.14)
! ! !
X 1 þ X 2 þ X 3
(6.15)
!
3
X ðt þ 1Þ¼
4.3.2.3 Proposed adaptive particleegray wolf optimization heuristic
The value of c 1 and c 2 , which are usually called “acceleration
coefficients,” are often set as constants, most likely c 1 ¼ c 2 ¼ 1
or c 1 ¼ c 2 ¼ 2. These values are found by empirical studies to
balance the cognitive and social components, which also balance
the exploration and exploration phases. In this study, we propose
a formula to change the acceleration coefficients in each iteration.
The new coefficients are calculated as follows:

t
f x
t k
1
c ¼ 1:2 (6.16)
f ðgBestÞ

f x t
t k
2
c ¼ 0:5 þ (6.17)
f ðgBestÞ
t

t
where c and c stand for the coefficients at iteration t; f x t is the
1 2 k
fitness of particle k at iteration t; and f ðgBestÞ is the swarm’s
global best fitness. The values of 1.2 and 0.5 are also found by
empirical studies. We also modify the formula for inertia as
follows:
ðmaxIter tÞ wMax wMin
þ wMin (6.18)
w t ¼
maxIter
Finally, we update the velocity and position of particles by the
following equations:
t
t
t

t
v tþ1 ¼ w v þ c rand pbest x t þ c rand gBest x t
k k 1 k k 2 k
(6.19)
t
x tþ1 ¼ x þ v t (6.20)
k k k
Senel et al. [20] provided a novel hybrid PSOeGWO by replac-
ing a particle of the PSO with a value being the mean of the three
best wolves’ positions. In this study, we introduce a probability
of mutation, which will trigger a small number of iterations of
GWO within the PSO main loop. The pseudocode for this algo-
rithm is given as follows: (Fig. 6.23).