Computer vision for solid waste sorting: A critical review of academic research

Area of Science:

• Environmental Science
• Computer Science
• Robotics

Background:

• Municipal solid waste (MSW) management increasingly relies on smart technologies like computer vision (CV) and robotics.
• The academic research in CV-enabled waste sorting is rapidly expanding, yet a comprehensive review of its evolution, current state, and future challenges is lacking.

Purpose of the Study:

• To critically review academic research on CV-enabled MSW sorting.
• To analyze prevalent CV algorithms, their performance, and research distribution across various waste streams and application domains.
• To identify research gaps and future directions for practical implementation.

Main Methods:

• Systematic literature review of academic research focused on CV-enabled MSW sorting.
• Comparative analysis of various CV algorithms, including traditional machine learning and deep learning approaches.
• Examination of research output distribution concerning waste sources, task objectives, application domains, and dataset availability.

Main Results:

• A significant trend towards adopting deep learning algorithms over traditional machine learning in CV for waste sorting.
• Enhanced robustness of CV in waste sorting due to advancements in computational power and algorithms.
• Uneven distribution of research across different sectors (household, commercial, construction) and a tendency towards simplified, artificial research environments and datasets.

Conclusions:

• Future research should prioritize real-world complexities and facilitate the industrial application of CV in waste sorting.
• Open sharing of waste image datasets is crucial for training and evaluating CV algorithms effectively.
• Continued advancements in CV algorithms and computational power are enhancing sorting efficiency and robustness.