Table 6.

Parameter setting of the baseline methods.

Baselines	Conv Layer	LSTM Layer	Attention Layer	Output Layer
LSTM	/	$\left[LSTM(128)\right]\times 1$	/	$\left[FC(128)\right]\times 1$
ABLSTM	/	$\left[LSTM(128)\right]\times 1$	$FC(128)\times 1$ tanh SoftMax	$\left[FC(128)\right]\times 1$
Deep-Conv LSTM	$\left[CNN(5),64\right]\times 4$	$\left[LSTM(128)\right]\times 2$	$FC(128)\times 2$ Sigmoid	$\left[FC(128)\right]\times 1$
TPN	$\left[\begin{array}{c}CNN(24),32\\ CNN(16),64\\ CNN(8),96\end{array}\right]$ Dropout = 0.1 Maxpool(8)	/	/	$\left[FC(96)\right]\times 1$
MSRLSTM	$\left[\begin{array}{c}CNN(3),64\\ CNN(2),128\\ CNN(2),128\end{array}\right]$ Res Maxpool(2)	$\left[LSTM(128)\right]\times 2$	$FC(128)\times 2$ SoftMax	$\left[\begin{array}{c}FC(256)\\ FC(512)\\ FC(1024)\end{array}\right]$

Note: CNN (a),(b) is convolutional layer, where a is the size of convolutional kernel, (b) is the kernel number; LSTM (c) is LSTM layer, where c is the size of hidden layer; FC (d) is fully connected layer, where d is the size of FC layer. Maxpool (e) is the max pooling layer, e is the size of pooling kernel; Res means a residual option after the previous convolutional layer.