Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

One-class anomaly detection through color-to-thermal AI for building envelope inspection

Polina Kurtser, Kailun Feng, Thomas Olofsson, Aitor De Andres

TL;DR

This work tackles the problem of detecting thermal anomalies in building envelopes without relying on labeled anomaly data. It introduces a one-class anomaly detection framework that learns a color-to-thermal mapping via a pix2pix-based network to predict thermal distributions from RGB imagery, and then flags anomalies as large mismatches between observed and predicted thermal maps. The model is trained on anomaly-free RGB-thermal pairs under selected outdoor conditions, enabling detection of thermal bridges when evaluating across different environmental scenarios. By reducing labeling needs and leveraging RGB-thermal alignment, the approach supports practical, scalable automatic inspection of large urban areas using mobile platforms.

Abstract

We present a label-free method for detecting anomalies during thermographic inspection of building envelopes. It is based on the AI-driven prediction of thermal distributions from color images. Effectively the method performs as a one-class classifier of the thermal image regions with high mismatch between the predicted and actual thermal distributions. The algorithm can learn to identify certain features as normal or anomalous by selecting the target sample used for training. We demonstrated this principle by training the algorithm with data collected at different outdoors temperature, which lead to the detection of thermal bridges. The method can be implemented to assist human professionals during routine building inspections or combined with mobile platforms for automating examination of large areas.

One-class anomaly detection through color-to-thermal AI for building envelope inspection

TL;DR

This work tackles the problem of detecting thermal anomalies in building envelopes without relying on labeled anomaly data. It introduces a one-class anomaly detection framework that learns a color-to-thermal mapping via a pix2pix-based network to predict thermal distributions from RGB imagery, and then flags anomalies as large mismatches between observed and predicted thermal maps. The model is trained on anomaly-free RGB-thermal pairs under selected outdoor conditions, enabling detection of thermal bridges when evaluating across different environmental scenarios. By reducing labeling needs and leveraging RGB-thermal alignment, the approach supports practical, scalable automatic inspection of large urban areas using mobile platforms.

Abstract

We present a label-free method for detecting anomalies during thermographic inspection of building envelopes. It is based on the AI-driven prediction of thermal distributions from color images. Effectively the method performs as a one-class classifier of the thermal image regions with high mismatch between the predicted and actual thermal distributions. The algorithm can learn to identify certain features as normal or anomalous by selecting the target sample used for training. We demonstrated this principle by training the algorithm with data collected at different outdoors temperature, which lead to the detection of thermal bridges. The method can be implemented to assist human professionals during routine building inspections or combined with mobile platforms for automating examination of large areas.
Paper Structure (13 sections, 3 equations, 3 figures, 2 tables)

This paper contains 13 sections, 3 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Introduction to relevant AI frameworks
  3. Generative-adversarial networks
  4. Image-to-image translation networks
  5. Color-to-thermal networks
  6. One-class classification
  7. Algorithm structure
  8. Implementation
  9. Data acquisition
  10. Preprocessing
  11. Network training
  12. Results and discussion
  13. Conclusion

Figures (3)

  • Figure 1: Algorithm workflow in black, with the corresponding human-driven steps in gray. After training A color-to-thermal GAN trained on data with the desired properties first produces the expected thermal distribution of a building envelope image (RGB). Then, the algorithm compares the expected and actual thermal distributions. Strong mismatches between both are labelled as anomalies.
  • Figure 2: Performance evaluation of the trained networks (a,b) winter4Net, (c,d) winter8Net, and (e,f) SummerNet when making predictions on their own test data sets. The histogram plots show the average pixel deviation for all the images in the corresponding test sets, with the red vertical lines indicating the deviation for the displayed images. Scale bars in $^o$C over the outdoors temperature.
  • Figure 3: Anomaly detection when using the (a,b) Winter4Net on Winter8 data, (c,d) SummerNet on winter4 data, and (e-f) SummerNet on winter4 data. The histogram plots show the average pixel deviation for all the images in the corresponding test sets, with the red vertical lines indicating the deviation for the displayed images. Scale bars in $^o$C over the outdoors temperature.