import torch.nn as nn
import torch.nn.functional as F
[docs]
class DCNN(nn.Module):
"""
A Deep Convolutional Neural Network (DCNN) for Human Activity Recognition (HAR)
using wearable sensors, based on the methodology presented in the paper
"Human Activity Recognition using Wearable Sensors by Deep Convolutional Neural Networks."
This DCNN architecture is designed to process 2D activity images generated from
wearable sensor signals and classify them into different activity categories.
"""
def __init__(self, num_classes=6):
"""
Initializes the DCNN with the given number of output classes.
Parameters:
-----------
num_classes: int, optional
The number of activity classes. Defaults to 6.
"""
super(DCNN, self).__init__()
self.conv1 = nn.Conv2d(in_channels=1, out_channels=5, kernel_size=5, stride=1, padding=0)
self.pool1 = nn.AvgPool2d(kernel_size=4, stride=4, padding=0)
self.conv2 = nn.Conv2d(in_channels=5, out_channels=10, kernel_size=5, stride=1, padding=0)
self.pool2 = nn.AvgPool2d(kernel_size=2, stride=2, padding=0)
self.fc1 = nn.Linear(in_features=10*2*6, out_features=120)
self.fc2 = nn.Linear(in_features=120, out_features=num_classes)
[docs]
def forward(self, x):
""" Defines the forward pass of the DCNN.
This method processes the input 2D activity images through a series of
convolutional and pooling layers, followed by fully connected layers
to produce the final class probabilities.
Parameters:
-----------
x: torch.Tensor
The input tensor representing the batch of 2D activity images. Shape should be (batch_size, 1, height, width).
Returns:
--------
torch.Tensor:
The output tensor containing the class probabilities for each input. Shape will be (batch_size, num_classes).
"""
x = self.pool1(F.elu(self.conv1(x)))
x = self.pool2(F.elu(self.conv2(x)))
x = x.view(-1, 10*2*6)
features = F.elu(self.fc1(x))
y = F.softmax(self.fc2(features), dim=1)
return y