PCB-Vision Dataset
Description:
The PCB-Vision dataset is a multiscene RGB-Hyperspectral benchmark dataset comprising 53 Printed Circuit Boards (PCBs). The RGB images are collected using a Teledyne Dalsa C4020 camera on a conveyor belt, while hyperspectral images (HSI) are acquired with a Specim FX10 spectrometer. The HSI data contains 224 bands in the VNIR range [400 - 1000]nm.
Data Format
RGB Images: .png files
PCB Masks: .jpg files
HSI Data: Each hyperspectral data cube is accompanied by a data file and a .hdr file.
Folder Organization
PCBVision
HSI/
53 subfolders (one for each PCB)
'General_masks' folder for 'General' segmentation ground truth
'Monoseg_masks' folder for 'Monoseg' segmentation ground truth
'PCB_Masks' folder for masks of the 53 PCBs in the hyperspectral cube
RGB/
53 .jpg images
'General' folder for RGB images 'General' segmentation ground truth
'Monoseg_masks' folder for RGB images 'Monoseg' segmentation ground truth
Data Classes in Masks
Masks (both 'General' and 'Monoseg') contain 1 to 4 segmentation classes:
0: "Others"
1: "IC"
2: "Capacitors"
3: "Connectors"
Code Repository
To facilitate reading and working with the data, Python codes are available on the GitHub repository:
https://github.com/hifexplo/PCBVision
Citation
If you use this dataset, please cite the following article:
Word:
Arbash, Elias, Fuchs, Margret, Rasti, Behnood, Lorenz, Sandra, Ghamisi, Pedram, & Gloaguen, Richard. (2024). PCB-Vision: A Multiscene RGB-Hyperspectral Benchmark Dataset of Printed Circuit Boards (Version 1) [Data set]. Rodare. http://doi.org/10.14278/rodare.2704
Latex:
@article{arbash2024pcb, title={PCB-Vision: A Multiscene RGB-Hyperspectral Benchmark Dataset of Printed Circuit Boards}, author={Arbash, Elias and Fuchs, Margret and Rasti, Behnood and Lorenz, Sandra and Ghamisi, Pedram and Gloaguen, Richard}, journal={arXiv preprint arXiv:2401.06528}, year={2024} }
Contact
For further information or inquiries, please visit our website:
https://www.iexplo.space/
Contact Email: e.arbash@hzdr.de
