LANITED\u2019s FR4 Inspection Dataset (LFR4ID)

The Laboratorio Nacional de Investigaci\u00f3n en Tecnolog\u00edas Digitales (LANITED) has developed a dataset of FR-4  copper boards captured during the PCB manufacturing process.This dataset contains images acquired directly from a PCB manufacturing cell using industrial cameras Genie Nano C4900 (4912 \u00d7 3684 pixels) and Genie Nano XL C5100 (5120 \u00d7 5120 pixels). Assisted by collaborative robots UR3 and UR5, variations in the position and orientation of the FR-4 boards were introduced. The images were captured during machining, component placement, and reflow processes. To enhance the robustness of the proposed dataset, the images were cropped to focus exclusively on the circuit region, removing background noise that could interfere with the training process. Subsequently, data augmentation techniques were applied, including lighting variations, rotation, and vertical flipping, in order to increase dataset diversity and generalization capability.The final dataset consists of 14 classes (ELL4LG, INDUCTOR 5.6UH, R-1812, DO-214AA, DO-214AB, SOD123W, C-1411 Case B, C-1210, HSOP8, drill misalignment, open circuit, etching failure, missing hole, nets) with a total of 25,040 images, divided into 60% for training, 20% for validation, and 20% for testing.