7A-3-4
(11)
where SM-EuM+EjfvL,) and
E,M=Ejh.Li)+En(hA)- £,,(£,,i,) and are
same as the terms in Equation (1). The energy relevant to
interlayers are
E\2 =(X \ X
El 2l =a 2 x
L )>\ *1 '2 k l
Ito
L J'l k 2 <1 k \
(12)
(13)
B l \kfjc2 is a connectivity variable from neuron (i v k { ) in layer
L, to neuron (i 2 ,k. 2 ) in layer L,. B 2 is a similar term. They
change dynamically during iterations. We also
have £ i2 , m * 2 ^B 2 \k^ because contributions from one layer
to another layer are non-symmetric. Using energy function
(11), we can recognize the trucks when a global minimized
energy value is achieved.
The connectivity term
Æ'Vi'A -<
<fv
-2xf
0
2x V2,^ — I if Line /c, 6 Area k 2 and i 2 = 0
(14)
if Line G Area & 2 aw/ / 2 = 1
otherwise
contributes when a model region is a truck top and an input line
belongs to an input region or when the model region is a truck
shadow and the input line belongs to an input region. Similarly,
B 2l i 2 kj t k\ is defined as
B~ i 2 k 2 i t k t —
2x
VI,.,*, — — I if Line fc, e Area k^andij =0
-2x| VI, > ~~
if Line e Area k^andi-, = 1
otherwise
(15)
The following describes the process of the two-layer Hopfield
neural network:
a) Calculate connectivity parameters C\, kjl of layer 1 and C2 ikj ,
of layer 2.
b) Set the initial states VI of layer 1 and V2 b of layer 2
'1*1 '2*2
using equation (11) respectively.
c) Obtain B l2 h k t i 2 k 2 using equation (14).
d) Update the values of III -. and Vf,
r 'i*i '1*1
for (Z, = 0; I, < m,; Zj ++)
for(k ] =0\k x <n x \ k l ++)
<*:=«!,•*, + ±(*,+2**2 +3*K 3 + K 4 )
K, =hxf(u i; iti )=hx(Aj j J j Cl,. iM ,V% -
B(Zv\ IJ -i)-c'Zvi lil -
D<L Vi t, - 1 )- £ I V ' 1 /., - 9 iXZ B WA V2 W,>
j 2**1 12 *2
*2 =hxf(u i; A + ±tf,)
^3 =/*X/(Ml! 1 * ) + 7*2>
^ 4 = /ix/( M i; i , i+ ^ 3 )
e) Compute Vl';| = g( M i;;|).
f) Obtain using equation (15).
g) Update the values of u2 . and V2 ,
'2*2 '2*2
for (i 2 = 0; Z 2 < ffl,; i 2 ++)
for (kj = 0; k 2 < n 2 ; &,++)
«2;;’ = < * 2 +j;( K] +2*K 2 +3*K 3 + K 4 ) Where
K, = hxf(u2' i2ki ) = ¿x(AXXC2, 2 * 2 ;,V2', -
-l)-C^V2 fj/ -
^1^*2 -5 2 XX fi '2*2',*/ 1 '-,*,)
j j*k 2 i, *,
K 2 =hxf{u2\ iki + ±JC,)
K 3 =hxf(u2'. ki +±K 2 )
K 4 =hxf(u2' iik2 + K 2 )
h) Update: V2^ = s(«2&) •
i) If converge, then stop, otherwise go back to step c.
Process of the two-layer Hopfield neural network