Wheat Impurity Image Dataset for Object Detection
Description
# Wheat Impurity Image Dataset for Object Detection This dataset provides high-resolution images of wheat kernels mixed with four common impurities (bark, stone, straw, wheat husk). Images were acquired dynamically using a prototype online monitoring platform under continuous material flow conditions. ## Dataset Details - **Camera:** Hikvision MV-CS060-10GC-PRO - **Image Resolution:** 3072 x 2048 (JPG) - **Annotation Format:** YOLO (.txt) ## Annotation Structure Each image in the `images/` folder has a corresponding `.txt` file in the `labels/` folder. Bounding box coordinates are normalized to [0, 1] following the standard YOLO format: `<class_id> <x_center> <y_center> <width> <height>` ## Classes The dataset contains 4 impurity categories. The mapping is: 0: bark 1: stone 2: straw 3: wheat husk ## Directory Layout - `images/`: Original image files. - `labels/`: Bounding box annotations. - `classes.txt`: Category names.
Files
Steps to reproduce
1.Hardware Setup: A prototype monitoring system was used, including a rotary quantitative feeder, a vibratory dispersion mechanism, and a Hikvision MV-CS060-10GC-PRO industrial camera. 2.Data Acquisition: Wheat kernels mixed with impurities were fed into the system. Images were captured dynamically under continuous material flow conditions at a flow rate of 75 mm/s. 3.Processing & Annotation: The original images (3072 x 2048 pixels) were captured in BMP format and subsequently converted to JPG to optimize storage while maintaining key visual features. All images were manually annotated using LabelImg software, following the standard YOLO format (.txt).
Institutions
- Henan University of TechnologyHenan, Zhengzhou