Cafeteria Tray Leftover Area Detection Image Dataset

With the rising demand for hygiene and efficiency among modern individuals, the automation of cafeteria cleaning has become a key development direction. However, current solutions for automated tray cleaning detection have limitations in recognition accuracy and real-time performance. Existing datasets are often limited to laboratory environments, lacking diversity in real usage scenarios. This dataset aims to enhance the recognition and adaptability of detection systems through rich image data. During data collection, we used high-resolution cameras to capture real tray images in various cafeteria environments to cover different lighting, angles, and types of tableware. Quality control ensures precision and reliability through multiple rounds of expert annotation and consistency review. The annotation team comprises image processing and machine learning experts. The data undergoes image enhancement and preprocessing steps, including normalization and segmentation, and is stored in a layered folder format as JPG files. This dataset is characterized by an annotation accuracy of over 95%, with consistency check results showing errors of less than 1%. The use of innovative hybrid augmentation techniques improves data diversity and generality, significantly enhancing the predictive accuracy of detection models. Compared to existing similar datasets, our data samples amount to 50,000, covering more real dining scenes, addressing the problem of insufficient generalization capability. For enhancing the accuracy of automated cleaning, it has important application value. Our dataset uniquely covers different lighting environments and weather effects, offering higher practicality and scalability.