Approach to image preprocessing for household waste classification in circular economy
Вантажиться...
Дата
2025
Назва журналу
Номер ISSN
Назва тому
Видавець
Анотація
This work proposes an intelligent image preparation pipeline for stable multi-class classification of household waste in field conditions, emphasizing data quality control rather than architectural complexity. A coordinated scheme of non-reference quality assessment, guided selection of training frames, and carefully designed partitioning protocols significantly improves model stability under realistic variations in lighting, background contamination, small object occlusions and inter-site distribution shifts. Using MobileNetV3-Small with standardized hyperparameters, the study demonstrates higher reliability, reduced metric variability, and consistent behavior across critical subsets without sacrificing class representativeness. The methodology is reproducible, hardware-efficient, and suitable for industrial sorting systems, supporting circular economy practices by increasing fraction purity and reducing material loss. Future work will extend cross-site benchmarking and explore threshold dynamics for different waste categories.
Опис
Ключові слова
Бібліографічний опис
Derzhak V.V., Mazurets O.V. Approach to image preprocessing for household waste classification in circular economy. Resource-Saving Technologies of Apparel, Textile & Food Industry. Proceedings of International Scientific and Practical Conference. November 20, 2025. Khmelnytskyi, Ukraine. Pp. 319-323.