Related Experiment Video
Updated: May 12, 2026
06:08
Quantitative Visualization and Detection of Skin Cancer Using Dynamic Thermal Imaging
Published on: May 5, 2011
16.8K
Automated detection and classification of concealed objects using infrared thermography and convolutional neural
WeeLiam Khor1,2, Yichen Kelly Chen3, Michael Roberts3,4
1Department of Mechanical Engineering Sciences, University of Surrey, Guildford, GU2 7XH, UK.
Scientific Reports
|April 9, 2024
Summary
Convolutional neural networks (CNNs) effectively classify infrared images for security scanning. Pre-processing thermal images with methods like Fuzzy-c clustering significantly improves concealed object detection accuracy.
Area of Science:
- Computer Vision
- Security Technology
- Machine Learning
Background:
- Infrared thermography offers non-invasive security scanning for concealed objects.
- Detecting heat signatures on clothing surfaces is challenging due to infrared's limited penetration.
- Automated classification of thermal images is crucial for effective security applications.
Purpose of the Study:
- To evaluate the effectiveness of a convolutional neural network (CNN) for classifying infrared images in security scanning.
- To investigate the impact of different image pre-processing techniques on CNN performance for concealed object detection.
- To determine the optimal pre-processing method for enhancing thermal image analysis.
Main Methods:
- Utilized the ResNet-50 convolutional neural network (CNN) architecture.
- Pre-trained the CNN model on the ImageNet database and fine-tuned it with experimental infrared images.
- Explored four image pre-processing approaches: raw infrared, region-of-interest (ROI) cropping, K-means clustering, and Fuzzy-c clustering.
- Evaluated model performance using the receiver operating characteristic (ROC) curve and calculating the area under the curve (AUC).
Main Results:
- The CNN model achieved varying performance levels based on image pre-processing.
- The Area Under the Curve (AUC) values were 0.8923 for raw images, 0.9256 for ROI cropped images, 0.9485 for K-means clustered images, and 0.9669 for Fuzzy-c clustered images.
- Fuzzy-c clustering yielded the highest detection accuracy, indicating its effectiveness in enhancing thermal image analysis.
Conclusions:
- Convolutional neural networks (CNNs) demonstrate significant potential for automated security scanning using infrared imagery.
- Image pre-processing techniques, particularly Fuzzy-c clustering, substantially improve the accuracy of concealed object detection.
- Removing irrelevant information and highlighting key features in thermal images enhances CNN prediction performance.
