PCBClustering: Image-Derived Morphometric Descriptors of Waste Printed Circuit Boards from the PCB-DSLR and CRUSADE Datasets

This dataset contains image-derived morphometric descriptors of waste printed circuit boards (WPCBs), generated using the PCBClustering application developed within the Horizon Europe CRUSADE project (Grant Agreement No. 101138642). The descriptors were extracted from conventional RGB photographs and are intended to support the visual characterisation, comparative analysis and unsupervised clustering of heterogeneous waste PCBs. Datasets The repository contains descriptors extracted from two image datasets: PCB-DSLR dataset: 165 images corresponding to 165 distinct waste PCBs, selected from the publicly available PCB-DSLR dataset. One image per physical board was used to avoid repeated observations of the same PCB. CRUSADE dataset: 42 images of automotive waste PCBs obtained from electronic units dismantled from end-of-life vehicles within the CRUSADE project. Image-derived descriptors Five morphometric descriptors were calculated using an automated computer-vision workflow combining U²-Net-based background removal and Segment Anything Model 3 (SAM 3) for component segmentation. Image calibration was used to express area measurements in physical units. The extracted descriptors are: pcbSize: Total projected area of the PCB (mm²). numberOfComponents: Number of detected electronic components on the PCB. areaOfComponents: Total projected area occupied by the detected components (mm²). numberOfIC: Number of detected components classified as integrated circuits through the segmentation workflow. areaOfIC: Total projected area of the detected integrated-circuit components (mm²). Repository contents Four comma-separated values (CSV) files are provided: extracted_features-trained-PCB-DSLR-3comp.csv — Three descriptors (pcbSize, numberOfComponents, areaOfComponents) for 165 PCB-DSLR images. extracted_features-trained-PCB-DSLR-5comp.csv — All five descriptors for the same 165 PCB-DSLR images. extracted_features-trained-CRUSADE-42PCBs-3comp.csv — Three descriptors for 42 CRUSADE automotive PCB images. extracted_features-trained-CRUSADE-42PCBs-5comp.csv — All five descriptors for the same 42 CRUSADE automotive PCB images. Each CSV record corresponds to an individual image identified by its filename. The three- and five-descriptor configurations represent alternative feature sets for the same images and are not independent observations. Intended use and limitations The datasets support the reproduction and further investigation of descriptor-based unsupervised clustering approaches for waste PCB characterisation. They can be used to compare clustering models trained with three or five descriptors, investigate differences between electronic-waste streams, and explore the potential of computer vision for preliminary PCB sorting. The descriptors quantify visible geometric and component-related characteristics. They are not direct measurements of metal concentrations, recoverable metal mass or chemical composition. In particular, integrated-circuit counts depend on segmentation-model predictions and may include other packaged semiconductor devices. The repository provides the extracted descriptor tables rather than the original PCB images, chemical characterisation measurements, trained machine-learning models or clustering assignments. These data accompany the manuscript PCBClustering: A Web-Based Platform for Vision-Driven Characterisation and Unsupervised Grouping of Printed Circuit Boards Powered by EosCloud. Funding: European Union Horizon Europe, CRUSADE project, Grant Agreement No. 101138642.