DefectoMCU: A Dataset for Microcontroller Board Defect Detection

Description

This dataset is a specialized annotated image dataset developed for microcontroller board damage detection using computer vision and deep learning. The dataset contains 7,169 images across 14 classes, covering both good-condition microcontroller boards and defective components. The classes include ATmega2560_Damage, ATmega328_Damage, Board_Damage, Boot_Button_Damage, Capacitor_Damage, Good_Arduino_Mega, Good_Arduino_Nano, Good_Arduino_Uno, Good_ESP32, Pin_Damage, Power_Port_Damage, Reset_Button_Damage, USB-B_Port_Damage, and USB_Input_Damage. The images represent real-world microcontroller board conditions with variations in illumination, board orientation, background, and image quality. All images are standardized to a resolution of 640 × 640 pixels and are provided with YOLO-compatible object-detection annotations. The dataset is organized into training, validation, and testing subsets. The dataset was prepared using Roboflow, including image preprocessing and augmentation applied to the training subset. The final dataset contains 6,273 training images, 597 validation images, and 299 test images. The training-set augmentation was performed after the original dataset split, with the original dataset containing 2,987 images before training-set augmentation. The dataset is provided to support reproducible research in microcontroller defect detection, object detection, automated visual inspection, and computer vision-based quality assessment. A metadata file, metadata_MicrocontrollerBoard.csv, is also provided to facilitate image-level analysis, dataset exploration, and further reuse of the dataset.

Steps to reproduce

The dataset was constructed for microcontroller board defect detection using computer vision. Images representing good-condition and defective microcontroller boards and components were collected and organized into 14 object-detection classes. The images were annotated for object detection and processed using Roboflow. Image preprocessing included auto-orientation and resizing to 640 × 640 pixels. The dataset was divided into training, validation, and test subsets before training-set augmentation. The original dataset contained 2,987 images, with augmentation applied to the training subset to produce the final dataset of 7,169 images, comprising 6,273 training images, 597 validation images, and 299 test images. The final dataset is provided in a YOLO-compatible format with corresponding image and annotation files. A metadata file (metadata_MicrocontrollerBoard.csv) is included for image-level analysis and dataset exploration.

Institutions

• United International University

  Dhaka Division, Dhaka

Categories

Related Links

Licence