YO-Dataset v1.0: A Modular Air-Cooling Equipment Fault Detection Dataset for Smart Substations

research gap of lacking standardized datasets for industrial fault detection in power system scenarios. The dataset provides a comprehensive foundation for evaluating lightweight end-to-end object detection models in complex industrial environments. ## Data Collection All images were collected from multiple actual smart substations in China, covering various typical industrial working conditions: - **Lighting conditions**: Strong light, backlight, low light, and normal indoor lighting - **Shooting angles**: Front view, side view, and oblique view - **Occlusion scenarios**: Cable occlusion, equipment overlap, and partial obstruction ## Dataset Statistics - **Total images**: 2,603 RGB images (640×640 resolution) - **Total annotated instances**: 10,007 - **Category distribution**: - Front plug-in board: 4,017 instances - Rear plug-in board: 4,688 instances - Fault indicator light: 1,302 instances - **Train-validation-test split**: 8:1:1 (2,141 training, 308 validation, 154 test images) ## Annotation Specification All images were manually annotated by professional power system engineers using LabelImg tool. The annotations follow the standard YOLO format, with each .txt file corresponding to an image file. Each line in the annotation file contains: `class_id x_center y_center width height` The three detection targets are defined as follows: 1. **Class 0: Front plug-in board** - Functional panel integrating control interfaces and status monitoring units on the front of the equipment 2. **Class 1: Rear plug-in board** - Physical support structure for backplane power supply bus and communication terminals 3. **Class 2: Fault indicator light** - Red light alarm device indicating abnormal status of the plug-in board ## File Structure