EE229-Project-Wireless-signal-classification

Introduction

This project shows the complete process from signal cutting to classification. Deep learning method(faster rcnn, yolov4) also supported.

File structure (traditional method)

stft_with_color.m : generate the spectrogram
recombination.py: adjusting the spectrogram
process.py: run segmentation and get the signal pieces
wash.m: zero-padding
hog_kmeans.m: K-Means clustering

Wireless Signal Auto-Classification Dataset(WSAD)

Number of data labels:

Class Wi-Fi BLE Others

Numbers 4018 1531 1715
Scale: 2k images
Some models' performance on this dataset:

Model mAP Inference Time

Faster RCNN 82.64% 57.7ms/img

YOLOv4 80.28% 5.6ms/img

You can download the whole dataset at https://jbox.sjtu.edu.cn/l/oFjsbK.

Reference

The main code of deep learning method references the Detectron2 platform and Darknet.