Centimetre radar reflectance signal for material classification
Data files
Mar 10, 2025 version files 226.23 MB
-
AllData.csv
166.30 MB
-
AllData.mat
59.93 MB
-
README.md
1.71 KB
Abstract
Detecting and classifying materials is crucial for many real-world applications such as industrial robots, and precision agriculture. In recent years, radar sensor-based material classification become more and more popular. In this work, we employ a portable centimeter-wave radar sensor to collect the reflectance signal from nine different kinds of materials.
Description of the data and file structure
Nine classes of materials were considered in this experiment. The signal from the empty container was considered as ‘Air’. In total, ten classes of materials were used, which are 1) Air, 2) Beans, 3) Dirt, 4) Flour, 5) Lentils, 6) Oats, 7) Rice, 8) River Rocks, 9) Sand, and 10) Sugar.
Each test material is placed in 3 containers, and the following procedure is used to collect data:
- Place the cup under the sensor. Make sure the cups are always placed in the same spot. This was done by placing the stand and cup on a sheet of paper. Trace the position of the stand legs and the cup which is centered under the stand
- Take 40 rounds of data collection for the first cup, as one round results in three scans
- After each round, shake the cup to change the material surface.
- Repeat the same steps from 1-3 for the second test cup.
-
Sample the third cup for 20 rounds with each round resulting in three scans. After each
round, shake the cup to change the material surface.
- Repeat steps 1-5 for each material except for the empty cup.
- Scan a single empty cup 15 times with each round also performing three scans.
Description of the Data
The data was averaged by three to reduce the noise, totally we have 956 observations(Some samples may get additional scans).
Two file types are provided: AllData.csv and AllData.mat.
AllData.csv is a table with dimensions 956 by 8193. Each row represents an observation, with the final column containing the label.
Conversely, AllData.mat is a struct containing the data in AllData.Data and the labels in AllData.Labels.
There are 956 observations in the dataset, categorized into 10 classes: 1) Air, 2) Beans, 3) Dirt, 4) Flour, 5) Lentils, 6) Oats, 7) Rice, 8) River Rocks, 9) Sand, and 10) Sugar.
For one observation, it contains 8119 data points.