This readme file was generated on [2026-03-9] by [Shaoyang peng] GENERAL INFORMATION Title of Dataset: Dataset and Code for RACON Thermographic Anomaly Detection Author Name:Shaoyang peng ORCID:0009-0009-8944-1518 Institution & Theme: Centre for Digital and Design Engineering Address: College Road Cranfield Bedfordshire MK43 0AL Email:slysp1@hotmail.com Author/Associate or Co-investigator Information Name: Haoxuan Deng ORCID:0009-0003-8819-8284 Institution & Theme: Centre for Digital and Design Engineering Address: College Road Cranfield Bedfordshire MK43 0AL Email: Haoxuan.Deng@cranfield.ac.uk Author//Principal Investigator/Supervisor Information Name: Sri Naga Pavan Addepalli ORCID:0000-0002-1466-7784 Institution & Theme: Centre for Digital and Design Engineering Address: College Road Cranfield Bedfordshire MK43 0AL Email: p.n.addepalli@cranfield.ac.uk Date of data collection: Geographic location of data collection:College Road Cranfield Bedfordshire MK43 0AL United Kingdom Information about funding sources that supported the collection of the data: The data were collected as part of a PhD research project conducted at Cranfield University. No specific external funding was received for the data collection. SHARING/ACCESS INFORMATION Licenses/restrictions placed on the data: CC-BY Links to publications that cite or use the data: Links to other publicly accessible locations of the data: Links/relationships to ancillary data sets: Was data derived from another source? If yes, list source(s): Recommended citation for this dataset: Peng, S., Deng, H., Addepalli, S., and Farsi, M., “RACON: Dataset and Code for Thermographic Anomaly Detection in CFRP,” Cranfield Online Research Data (CORD), 2026. DATA & FILE OVERVIEW File List:RACON ├── racon.py ├── racon.ipynb ├── dataset.py ├── dataset ├── originalData ├── train ├── test ├── val └── annotatedData ├── train └── test ├── val Additional related data collected that was not included in the current data package: Are there multiple versions of the dataset? no METHODOLOGICAL INFORMATION Description of methods used for collection/generation of data:The thermographic images used in this dataset originate from the publicly available Thermal Inspection Dataset for Defect Segmentation in CFRP Laminates published by Garcia Vargas and Fernandes (2025) on Mendeley Data (DOI: 10.17632/jrsb4b9yy5.1). The original dataset was acquired using a mid-wave infrared (MWIR) camera during active thermography experiments on carbon fibre reinforced polymer (CFRP) laminates with artificially embedded defects of different sizes and depths. In this repository, the original thermographic images were preprocessed and reorganised to support anomaly detection experiments. The dataset includes resized thermographic images and corresponding binary masks indicating defect regions. These annotations were derived from the original dataset and adapted for evaluation of anomaly localisation methods. In addition to the processed dataset, this repository provides the implementation of the RACON (Random Convolution) anomaly detection framework used in the associated publication. The code includes data preprocessing, multi-scale feature extraction using random convolution kernels, and KNN-based anomaly scoring for thermographic defect detection. Methods for processing the data: First, the original thermographic frames were converted into a unified image format and resized to a consistent spatial resolution suitable for model evaluation. Corresponding defect annotations were represented as binary masks, where pixels labelled as 0 denote defect-free regions and pixels labelled as 1 represent defect areas. For the RACON anomaly detection framework, the images were further processed through a preprocessing pipeline including Gaussian smoothing and multi-scale pooling to construct feature representations at different spatial resolutions. Random convolution kernels were then applied to extract feature maps, which were subsequently used to compute patch-level descriptors. Anomaly scores were finally obtained using a K-nearest neighbour (KNN) distance metric in the feature space. These preprocessing steps allow the dataset to be directly used for benchmarking thermographic anomaly detection methods. Instrument- or software-specific information needed to interpret the data: Requirements Python >= 3.10 Pytorch >= 2.0 It is recommnded to install the requirements using the following command: conda create -n racon python=3.10 conda activate racon pip install torch==2.2.0 torchvision==0.17.0 torchaudio==2.2.0 --index-url https://download.pytorch.org/whl/cu121 Standards and calibration information, if appropriate: Environmental/experimental conditions: The inspection setup was designed to highlight subsurface defects embedded within the laminate structure. Experimental conditions, including excitation configuration and acquisition parameters, are documented in the original dataset publication. Describe any quality-assurance procedures performed on the data: To ensure data consistency and reliability for anomaly detection experiments, several preprocessing and verification steps were performed. First, the thermographic images and corresponding binary masks were checked for completeness and correct alignment. Image formats were standardised and resized to ensure consistent spatial resolution across the dataset. The defect masks were verified to confirm that pixel-level annotations correctly correspond to the defect regions provided in the original dataset. Additional validation was performed by visual inspection of randomly sampled image–mask pairs to confirm annotation consistency and data integrity. People involved with sample collection, processing, analysis and/or submission: Sample collection Garcia Vargas, Iago; Fernandes, Henrique – Original dataset acquisition and publication. Data processing and analysis Shaoyang Peng – Dataset preprocessing, algorithm development (RACON), experimental evaluation, and repository preparation. Haoxuan Deng – Methodological support and implementation assistance. Sri Addepalli – Research supervision and methodological guidance. Maryam Farsi – Research supervision and review. DATA-SPECIFIC INFORMATION FOR:Racon Number of variables: 3 (RGB channels per image) Number of cases/rows: 1,045 images Missing data codes: No missing data are present in the dataset. Specialized formats or other abbreviations used: PNG – Portable Network Graphics image format used for thermographic images. py,ipynb- python file,Random Convolution-based anomaly detection framework used for experiments