demo-model v1 - Create impulse

Pertti Arvola / demo-model v1 Public

Target: Arduino Nano 33 BLE Sense (Cortex-M4F 64MHz) Clone this project

Impulse #1

An impulse takes raw data, uses signal processing to extract features, and then uses a learning block to classify new data.

Description	Recommended
Time series data Officially Supported Operates on time series sensor data like vibration or audio data. Lets you slice up data into windows.
Images Officially Supported Processes discrete images for object detection or classification.
Pre-processed features Officially Supported Processes pre-processed features, or non time-series data.

Did you know? You can bring your own DSP code.

Description	Author	Recommended
Flatten Officially Supported Flatten an axis into a single value, useful for slow-moving averages like temperature data, in combination with other blocks.	Edge Impulse
Spectral Analysis Officially Supported Great for analyzing repetitive motion, such as data from accelerometers. Extracts the frequency and power characteristics of a signal over time.	Edge Impulse
Spectrogram Officially Supported Extracts a spectrogram from audio or sensor data, great for non-voice audio or data with continuous frequencies.	Edge Impulse
IMU (Syntiant) Officially Supported Syntiant only. Great for analyzing repetitive motion, such as data from accelerometers. Extracts the frequency and power characteristics of a signal over time.	Syntiant
HR and HRV features Officially Supported Process PPG or ECG data into heart rate and heart rate variability features. This block is available for testing but requires an Enterprise license to deploy.	Edge Impulse
Raw Data Officially Supported Use data without pre-processing. Useful if you want to use deep learning to learn features.	Edge Impulse
Flatten Officially Supported Flatten an axis into a single value, useful for slow-moving averages like temperature data, in combination with other blocks.	Edge Impulse
Image Officially Supported Preprocess and normalize image data, and optionally reduce the color depth.	Edge Impulse
Audio (MFCC) Officially Supported Extracts features from audio signals using Mel Frequency Cepstral Coefficients, great for human voice.	Edge Impulse
Audio (MFE) Officially Supported Extracts a spectrogram from audio signals using Mel-filterbank energy features, great for both voice and non-voice audio.	Edge Impulse
Spectral Analysis Officially Supported Great for analyzing repetitive motion, such as data from accelerometers. Extracts the frequency and power characteristics of a signal over time.	Edge Impulse
Spectrogram Officially Supported Extracts a spectrogram from audio or sensor data, great for non-voice audio or data with continuous frequencies.	Edge Impulse
Audio (Syntiant) Officially Supported Syntiant only. Compute log Mel-filterbank energy features from an audio signal.	Syntiant
IMU (Syntiant) Officially Supported Syntiant only. Great for analyzing repetitive motion, such as data from accelerometers. Extracts the frequency and power characteristics of a signal over time.	Syntiant
HR and HRV features Officially Supported Process PPG or ECG data into heart rate and heart rate variability features. This block is available for testing but requires an Enterprise license to deploy.	Edge Impulse
Raw Data Officially Supported Use data without pre-processing. Useful if you want to use deep learning to learn features.	Edge Impulse

Did you know? You can bring your own model in PyTorch, Keras or scikit-learn.

Description	Author	Recommended
Classification Officially Supported Learns patterns from data, and can apply these to new data. Great for categorizing movement or recognizing audio.	Edge Impulse
Regression Officially Supported Learns patterns from data, and can apply these to new data. Great for predicting numeric continuous values.	Edge Impulse
Anomaly Detection (GMM) Professional Enterprise Find outliers in new data. A Gaussian mixture model (GMM) models the shape of data using a probability distribution. New data that is unlikely according to this model can be considered anomalous.	Edge Impulse
Anomaly Detection (K-means) Officially Supported Find outliers in new data. Good for recognizing unknown states, and to complement classifiers. Works best with low dimensionality features like the output of the spectral features block.	Edge Impulse
Visual Anomaly Detection - FOMO-AD Enterprise Detect visual anomalies. Extracts visual features using a pre-trained backbone, and applies a scoring function to evaluate how anomalous a sample is by comparing the extracted features to the learned model. Does not require anomalous data.	Edge Impulse
Classification Officially Supported Learns patterns from data, and can apply these to new data. Great for categorizing movement or recognizing audio.	Edge Impulse
Transfer Learning (Images) Officially Supported Fine tune a pre-trained image classification model on your data. Good performance even with relatively small image datasets.	Edge Impulse
Object Detection (Images) Officially Supported Fine tune a pre-trained object detection model on your data. Good performance even with relatively small image datasets.	Edge Impulse
Regression Officially Supported Learns patterns from data, and can apply these to new data. Great for predicting numeric continuous values.	Edge Impulse
Transfer Learning (Keyword Spotting) Officially Supported Fine tune a pre-trained keyword spotting model on your data. Good performance even with relatively small keyword datasets.	Edge Impulse
Anomaly Detection (GMM) Professional Enterprise Find outliers in new data. A Gaussian mixture model (GMM) models the shape of data using a probability distribution. New data that is unlikely according to this model can be considered anomalous.	Edge Impulse
Anomaly Detection (K-means) Officially Supported Find outliers in new data. Good for recognizing unknown states, and to complement classifiers. Works best with low dimensionality features like the output of the spectral features block.	Edge Impulse
Visual Anomaly Detection - FOMO-AD Enterprise Detect visual anomalies. Extracts visual features using a pre-trained backbone, and applies a scoring function to evaluate how anomalous a sample is by comparing the extracted features to the learned model. Does not require anomalous data.	Edge Impulse

Time series data

Input axes (3)

x, y, z

Window size

Window increase (stride)

Frequency (Hz)

Invalid frequency

Zero-pad data

Spectral Analysis

Name

Input axes (3)

Classification

Name

Input features

Spectral features

Output features

3 (horizontal, stay, vertical)

Output features

3 (horizontal, stay, vertical)

Time series data

Spectral Analysis

Classification

Output features

No data collected yet