A Novel Human Respiration Pattern Recognition Using Signals of Ultra-Wideband Radar Sensor

. 2019 Jul 30;19(15):3340. doi: 10.3390/s19153340

Algorithm 1 Algorithm for finding optimal convolutional layer depth parameter (

D_{o p t}

)

(1)
Set the Convolutional Layer Depth parameter ( $C L D_{n}$ ).
$C L D_{n}, (C L D_{n} \in C L D, n = 1, 2, 3, \dots, T) w h e n C L D = {1, 2, 3, \dots, T}$
(2)
Set the Kernel Size ( $K S_{j}$ ).
$K S_{j}, (K S_{j} \in K S, j = 1, \dots, 10) w h e n K S = {5, 9, 13, 17, 21, 25, 29, 33, 37, 41}$
(3)
Set the Kernel Count ( $K C_{k}$ ).
$K C_{k}, (K C_{k} \in K C, k = 1, \dots, 5) w h e n K C = {32, 64, 128, 256, 512}$
(4)
Set the Dense Layer Neuron Count ( $D L N C_{l}$ ).
$D L N C_{l}, (D L N C_{l} \in D L C N, l = 1, \dots, 4) w h e n D L N C = {256, 512, 1024, 2048}$
(5)
Define the Convolution Layer parameter combination ( $C o n v_{n - t h}$ , $n = 1, \dots, T$ ).

$⊙$ : combination for all cases in set of elements.
$C o n v_{n - t h} : {K S_{j} ⊙ K C_{k}}_{j = 1 \dots 10; k = 1 \dots 5}$
(6)
Define the Dense Layer parameter ( $D e n s e_{m - t h}$ , $m = 1, 2$ ).
$D e n s e_{m - t h} : {D L N C_{l}}_{j = 1 \dots 10; k = 1 \dots 5}$
(7)
Define the $P a r a m_{n - t h}$ by the combination ( $⊙$ ) of the parameters (1), (5) and (6) in Convolutional Layer depth $n$ .
$P a r a m_{n - t h} : {C L D_{n} ⊙ C o n v_{n - t h} ⊙ D e n s e_{m - t h}}_{n = 1, 2, 3; m = 1, 2}$
(8)
The recognition rate for each parameter combination and the average recognition rate at Convolutional Layer Depth n are calculated as follows.

Begin Loop:
for (n = 1; n < = T; n++)
- ①
  Configures z parameter combinations $C P_{n - t h}$ (Combination of Parameters) for all cases on depth n.
  $C P_{n - t h} : {P a r a m_{n - t h}^{1}, \dots, P a r a m_{n - t h}^{z}}$
- ②
  Configures the $A c c_{n - t h}$ by calculating the recognition rate of each
  $P a r a m_{n - t h}^{p} (p = 1, \dots z) for \dots C P_{n - t h} .$
  
  $A c c_{n - t h} : {a c c (P a r a m_{n - t h}^{1}), \dots, a c c (P a r a m_{n - t h}^{z})}$
- ③
  Compute average recognition rate $μ A c c_{n - t h}$ and store the result value and depth value n to $R e s u l t_{n}$ .
  $μ A c c_{n - t h} = \frac{1}{p} {\sum_{p = 1}^{z} a c c (P a r a m_{i - t h}^{p})}$
  
  $R e s u l t_{n} : {n, μ A c c_{n - t h}}$
  End Loop:
(9)
Find the maximum $μ A c c_{n - t h}$ of T Results and set the depth value n to the D_opt.
$D_{o p t} = {n | n \in R e s u l t_{n}, a r g m a x (μ A c c_{n - t h})}$