Waste Classification Recyclables and Domestic Waste Classification Dataset

* This dataset supports online use.Click here to jump.

The dataset contains 15,000 images (256×256 pixels each) covering a variety of recyclable materials, general waste, and household items in 30 different categories. With 500 images per category and 250 images per subcategory, the dataset provides a rich and diverse resource for research and development in the field of waste sorting and recycling. By providing a large number of high-quality images, the dataset aims to support the creation of robust and accurate waste sorting and classification systems.

Dataset structure

The dataset is organized in a hierarchical folder structure to ensure easy navigation and access. The main folder is named “image” and contains subfolders representing specific waste categories or items.These subfolder names serve as labels for their respective categories.Make it easy for researchers and developers to identify and exploit images for their specific needs.

You have to manually split the dataset into test, training, and validation.See alsoRecyclable and household waste sorting codesfor an example of how to do this.

Within each category subfolder, there are two different folders:

1. default: This folder contains standard or studio-quality images of waste objects. These images clearly and controllably represent the objects and can be used for initial training and testing of waste classification models. Each "default" subfolder contains 250 images.
2. real_world: This folder contains images of waste items in real-world scenes or environments. These images capture items in various situations, such as trash bins, on the ground, or in cluttered environments. Realistic images are critical for evaluating the performance and robustness of waste classification models in practical settings. Each "real_world" subfolder also contains 250 images.

All images in the dataset are provided in PNG format, ensuring high quality and compatibility with various image processing and machine learning libraries.

Waste Type

The dataset covers a wide range of waste categories and items, including:

• plastic:This category includes images of plastic water bottles, soda bottles, detergent bottles, shopping bags, trash bags, food containers, disposable cutlery, straws, and cup lids. These items make up a large portion of the plastic waste generated by households and are critical to recycling efforts.
• Paper and cardboard:This category includes images of newspapers, office paper, magazines, cardboard boxes and cardboard packaging. These items are often recyclable and play a vital role in reducing deforestation and conserving natural resources.
• Glass:This category includes images of beverage bottles, food jars, and cosmetic containers made of glass. Glass is a highly recyclable material and correct sorting and organization is essential for an efficient recycling process.
• Metal:This category includes images of aluminum soda cans, aluminum food cans, steel food cans, and aerosol cans. Metal waste has recycling value and can be processed efficiently if properly identified and separated.
• Organic waste:This category includes images of food waste, such as fruit peels, vegetable scraps, eggshells, coffee grounds, and tea bags. Organic waste can be composted or used to produce biogas, which reduces the burden on landfills and generates a valuable resource.
• textile:This category includes images of clothing and shoes. Textile waste is a growing concern and proper sorting can help with recycling and reduce the environmental impact of the fashion industry.

Please refer to the individual subfolders within the dataset for specific examples and instances of each waste category.

Use and application of datasets

The Recyclable and Household Waste Sorting Dataset offers a wide range of possibilities for researchers, engineers, and environmental enthusiasts. Some potential uses and applications of this dataset include:

• Develop and train machine learning models to automatically sort and categorize waste. The diverse images and real-world scenarios in the dataset enable the creation of robust and accurate classification models that can be deployed in waste management facilities, recycling centers, and smart waste bins.
• Analyze the visual characteristics and properties of different wastes. Researchers can use the dataset to study the unique visual attributes of various wastes, such as color, shape, texture, and size. This analysis can help develop more efficient and targeted waste sorting algorithms.
• Compare the performance of trash sorting algorithms on default and real images. This dataset allows researchers to evaluate and benchmark the accuracy and robustness of their algorithms in a controlled and real-world environment. This comparison can help identify the strengths and weaknesses of different approaches and guide the development of more reliable trash sorting systems.
• Study the impact of real-world scenarios on waste identification accuracy. The dataset contains realistic images, allowing researchers to study how factors such as lighting conditions, object occlusion, and background clutter affect the performance of waste classification models. This research can promote the development of more resilient and adaptive algorithms to cope with the challenges encountered in real-world waste management scenarios.