deep-learning-HAR-master
所属分类：人工智能/神经网络/深度学习
开发工具：Python
文件大小：56461KB
下载次数：40
上传日期：2018-04-04 15:21:37
上 传 者：
ant_me
说明： 一份用tensorflow平台做的cnn分类时序信号，是分类UCI 项目中的人体活动识别（HAR）数据集。该数据集包含原始的时序数据和经预处理的数据（包含 561 个特征）
(A CNN classification timing signal made by tensorflow platform is a human activity recognition (HAR) dataset in the classified UCI project. The dataset contains original time series data and preprocessed data (including 561 characteristics).)
文件列表:
utilities.py (1820, 2017-09-22)
data\.DS_Store (6148, 2015-02-15)
data\activity_labels.txt (80, 2012-10-10)
data\features.txt (15785, 2012-10-11)
data\features_info.txt (2809, 2012-10-15)
data\test\Inertial_Signals\body_acc_x_test.txt (6041350, 2012-11-29)
data\test\Inertial_Signals\body_acc_y_test.txt (6041350, 2012-11-29)
data\test\Inertial_Signals\body_acc_z_test.txt (6041350, 2012-11-29)
data\test\Inertial_Signals\body_gyro_x_test.txt (6041350, 2012-11-29)
data\test\Inertial_Signals\body_gyro_y_test.txt (6041350, 2012-11-29)
data\test\Inertial_Signals\body_gyro_z_test.txt (6041350, 2012-11-29)
data\test\Inertial_Signals\total_acc_x_test.txt (6041350, 2012-11-29)
data\test\Inertial_Signals\total_acc_y_test.txt (6041350, 2012-11-29)
data\test\Inertial_Signals\total_acc_z_test.txt (6041350, 2012-11-29)
data\test\subject_test.txt (7934, 2012-11-29)
data\test\X_test.txt (26458166, 2012-11-29)
data\test\y_test.txt (5894, 2012-11-29)
data\train\Inertial_Signals\body_acc_x_train.txt (15071600, 2012-11-29)
data\train\Inertial_Signals\body_acc_y_train.txt (15071600, 2012-11-29)
data\train\Inertial_Signals\body_acc_z_train.txt (15071600, 2012-11-29)
data\train\Inertial_Signals\body_gyro_x_train.txt (15071600, 2012-11-29)
data\train\Inertial_Signals\body_gyro_y_train.txt (15071600, 2012-11-29)
data\train\Inertial_Signals\body_gyro_z_train.txt (15071600, 2012-11-29)
data\train\Inertial_Signals\total_acc_x_train.txt (15071600, 2012-11-29)
data\train\Inertial_Signals\total_acc_y_train.txt (15071600, 2012-11-29)
data\train\Inertial_Signals\total_acc_z_train.txt (15071600, 2012-11-29)
data\train\subject_train.txt (20152, 2012-11-29)
data\train\X_train.txt (66006256, 2012-11-29)
data\train\y_train.txt (14704, 2012-11-29)
HAR-CNN.ipynb (282198, 2018-04-03)
data\test\Inertial_Signals (0, 2012-11-29)
data\train\Inertial_Signals (0, 2012-11-29)
data\test (0, 2018-04-03)
data\train (0, 2018-04-03)
data (0, 2015-02-15)
explore_data.ipynb (36748, 2018-04-03)
# Human Activity Recognition (HAR)
In this part of the repo, we discuss the human activity recognition problem using deep learning algorithms
and compare the results with standard machine learning algorithms that use engineered features.
The data can be downloaded from the [UCI repository](https://archive.ics.uci.edu/ml/datasets/human+activity+recognition+using+smartphones).
Also see G.Chevalier's [repo](https://github.com/guillaume-chevalier/LSTM-Human-Activity-Recognition) and A. Saeed's [blog](https://aqibsaeed.github.io/2016-11-04-human-activity-recognition-cnn/) where I have got lots of isnpiration.
## Contents
The `utils` folder contains the code for reading and processing the data into a tensor form. The generated tensors
has the dimensions
```
(batch, seq_len, n_channels)
```
where `batch` is the number of training examples in each batch, `seq_len` is the number of steps in the time series (128) and
`n_channels` is the number of channels where observations are made (9).
The aim is to clasify the activities correctly, which are
```
1 WALKING
2 WALKING_UPSTAIRS
3 WALKING_DOWNSTAIRS
4 SITTING
5 STANDING
6 LAYING
```
Below are the architectures used for training
Notebook | Description
-------- | ------
[explore_data](https://github.com/bhimmetoglu/seizure-forecast/blob/master/HAR/explore_data.ipynb) | Data exploration
[HAR-LSTM](https://github.com/bhimmetoglu/seizure-forecast/blob/master/HAR/HAR-LSTM.ipynb) | LSTM network
[HAR-CNN](https://github.com/bhimmetoglu/seizure-forecast/blob/master/HAR/HAR-CNN.ipynb) | Convolutional neural network(CNN)
[HAR-CNN-LSTM](https://github.com/bhimmetoglu/seizure-forecast/blob/master/HAR/HAR-CNN-LSTM.ipynb) | CNN + LSTM hybrid
[HAR-CNN-Inception](https://github.com/bhimmetoglu/seizure-forecast/blob/master/HAR/HAR-CNN-Inception.ipynb) | CNN with inception module
## Results
Method | Test accuracy
------ | -------------
CNN | 93%
LSTM | 88%
CNN+LSTM | 88%
CNN+Inception | 89%
[Xgboost](https://rpubs.com/burakh/har_xgb) | 96%
## CNN architecture
## LSTM architecture
近期下载者：
相关文件：
收藏者：