BioMedical Engineering OnLine

Table 6 Architecture of the 1D-CNN-LSTM model

From: The effectiveness of simple heuristic features in sensor orientation and placement problems in human activity recognition using a single smartphone accelerometer

Parts of architecture	Components of each part
Parts of architecture	Layer’s name	Number of filters	Kernel size	Pool size	Activation function	Padding type	Dropout ratio
CNN	Convolution	512	5		relu	Same
	Dropout						0.3
	Average pooling			3		Same
	Convolution	256	3		relu	Same
	Dropout						0.3
	Convolution	64	3		relu	Same
	Average pooling			3		Same
	Convolution	128	3		relu	Same
	Convolution	256	5		relu	Same
	Dropout						0.3
	Convolution	512	7		relu	Same
	Dropout						0.3
	Average pooling			3		Same

	Layer’s name			Number of units			Activation function
LSTM	LSTM			512			tanh

	Layer’s name			Number of neurons			Activation function
Fully connected network	Dense			100			relu
	Dense			28			relu
	Dense			64			relu
	Dense			6			softmax

Back to article page

ISSN: 1475-925X

Contact us

Submission enquiries: journalsubmissions@springernature.com