Multi-sensor datasets of ultrasonic and mmWave reflected signals for material classification

Our plain-English summary of Multi-sensor datasets of ultrasonic and mmWave reflected signals for material classification, by M.H. Sadiq, S.A. Abdulkareem, A.B. Al-Khalil (Data in Brief, 2026).

The study

• Authors: M.H. Sadiq, S.A. Abdulkareem, A.B. Al-Khalil
• Journal: Data in Brief
• Published: 2026;66:112760
• Read it: PubMed · Publisher · Free full text

Abstract

This article details a publicly accessible dataset of reflected signals from ultrasonic and millimetre-wave (mmWave) sensors used for material classification. The dataset was gathered using two sensing methods: the URM09 ultrasonic sensor operating at 40 kHz and the DFRobot mmWave C4001 radar sensor operating at 24 GHz. Six common materials were chosen for data collection: wood, plastic, metal, glass, cardboard, and asbestos. Each material was tested at three thicknesses and one size (25 cm × 25 cm). All measurements were taken from a fixed distance of 55 cm in a controlled indoor setting with foam-lined enclosures to reduce external noise. The raw analogue signals were processed with Savitzky-Golay filtering and sliding window techniques. Time-domain features (mean, standard deviation, energy, RMS, kurtosis, skewness) and frequency-domain features (spectral centroid, spectral bandwidth, dominant frequency) were extracted from each sample. For URM09 the dataset includes 1789 refined data points across all six material-thickness combinations and 565 records for six material-only classification. On the other hand, For C4001 the dataset includes 1504 refined data points across all five material-thickness combinations and 470 records for five material-only classification. This dataset supports research in robotics, industrial inspection, non-destructive testing, and sensor fusion applications where contactless material identification is essential.

© 2026 The Authors.

In plain English

This was a data article, not a health study. The authors, at the University of Duhok in Iraq, described a public dataset of signals reflected from samples of six materials: wood, plastic, metal, glass, cardboard and asbestos. Each material was prepared at 25 cm by 25 cm in three thicknesses. The asbestos samples were 0.1, 0.2 and 0.3 cm thick. The article did not say what type of asbestos product or fiber the samples were. A 40 kHz ultrasonic sensor and a 24 GHz millimetre-wave radar sensor were aimed at each sample from 55 cm away, inside a foam-lined enclosure. The radar sensor could not interact with asbestos, so asbestos data came only from the ultrasonic sensor. In a computer classification test, the ultrasonic sensor identified the six materials with 100% accuracy and the radar sensor identified five materials with 99%. Both results used only each material's most reliably distinguishable thickness, chosen from earlier tests. When the ultrasonic sensor had to tell the three asbestos thicknesses apart, accuracy was 64%. The authors noted that results may differ outside their controlled indoor setup.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Data article describing a sensor signal dataset, with a baseline machine-learning (Random Forest, stratified 10-fold cross-validation) check of how well the materials could be told apart
• Who was studied: No people were studied. Samples of six materials (wood, plastic, metal, glass, cardboard, asbestos) were prepared at 25 cm by 25 cm, each in three thicknesses, and measured in a laboratory at the Department of Computer Science, University of Duhok, Iraq. The asbestos samples were 0.1 cm, 0.2 cm and 0.3 cm thick. The type of asbestos was not stated.
• Exposure: Not a human asbestos exposure. Samples were measured with a URM09 ultrasonic sensor emitting 40 kHz sound waves and a DFRobot C4001 mmWave radar sensor emitting 24 GHz waves. Both sensors were at a fixed distance of 55 cm in a foam-lined enclosure.
• Main results:
  • Material identification regardless of thickness: the URM09 ultrasonic sensor reached 100% accuracy for 6 materials, and the C4001 radar sensor reached 99% for 5 materials.
  • The material-only results used a reduced dataset. After examining earlier classification results, the authors selected the thickness level that gave the most reliable and discriminative response for each material.
  • When materials and their three thickness levels had to be identified together, accuracy was 72% for the URM09 sensor (6 materials) and 77% for the C4001 sensor (5 materials).
  • Telling apart the three thicknesses of asbestos with the URM09 ultrasonic sensor reached 64% accuracy, the lowest of the six materials. Cardboard reached 68%.
• The authors' conclusion: The authors concluded that the two sensors showed complementary strengths, which could be combined in sensor-fusion research to make material classification more robust.
• Limitations the authors state:
  • The mmWave C4001 sensor could not interact with asbestos, so asbestos data exist only for the URM09 ultrasonic sensor.
  • All data were collected at a fixed 55 cm distance in a controlled indoor enclosure. Signals may differ in open environments, at longer distances or under other conditions.
  • Both sensors have wide fields of view. In uncontrolled settings, stray reflections and background interference may affect performance.
  • The 12-bit analogue-to-digital converter may introduce quantisation effects for applications needing higher precision.

Related studies

• Signal Detection by Sensors and Determination of Friction Coefficient During Brake Lining Movement — L. Hrabovský, V. Molnár, et al. (2024)
• Exploring the Impact of DNA Methylation on Gene Expression in CRC: A Computational Approach for Identifying Epigenetically Regulated Genes in Multi-Omic Datasets — A.S. Blindu, S. Berardelli, et al. (2026)
• Label-free machine learning detects a multi-cancer signal in a noninvasive urine and plasma metabolite panel: a case-control study — D.P. Patel, S. Chung, et al. (2026)
• Asbestos containing materials detection and classification by the use of hyperspectral imaging — G. Bonifazi, G. Capobianco, et al. (2018)
• Chromosomes and chromosomal evolution in human mesotheliomas as reflected in sequential analyses of two cases — G. Stenman, K. Olofsson, et al. (1986)
• Ultrasonic manifestations of diffuse pericardial mesothelioma masquerading as pericardial effusion: a case report — S. Zhu, Y. Zhou (2026)

Browse return to medical research.