The rise of LEAN and small-batch manufacturing calls for ๐ณ๐น๐ฒ๐ ๐ถ๐ฏ๐น๐ฒ ๐ฎ๐๐๐ผ๐บ๐ฎ๐๐ฒ๐ฑ ๐๐ผ๐ฟ๐ธ๐๐๐ฎ๐๐ถ๐ผ๐ป๐ that can seamlessly switch between different waste types over time. Our latest study explored deep learning models from the Segment Anything Model (SAM) family for extracting waste objects from raw images, followed by a classification stage to ensure accurate waste sorting. This two-step procedure eliminates the need for dedicated waste detection and segmentation algorithms, enhancing robotic waste sorting by efficiently segmenting and classifying highly variable objects.
Tested on ๐ณ๐ผ๐๐ฟ ๐๐๐ฒ ๐ฐ๐ฎ๐๐ฒ๐ (floating waste, municipal waste, e-waste and smart bins), ๐ผ๐๐ฟ ๐บ๐ฒ๐๐ต๐ผ๐ฑ ๐ฎ๐ฐ๐ต๐ถ๐ฒ๐๐ฒ๐ฑ ๐ด๐ฒ-๐ต๐ณ% ๐ฎ๐ฐ๐ฐ๐๐ฟ๐ฎ๐ฐ๐, demonstrating its robustness and industrial applicability.
By simplifying deployment, improving efficiency, and reducing costs, this approach has the potential to enhence robotic waste sorting and boost recycling and material utilization in manufacturing.
You can explore the full study here
๐๐๐๐ต๐ผ๐ฟ๐: Arso Vukicevic, Milos Petrovic, Nebojลกa Juriลกeviฤ, Djapan Marko PhD, Nikola Kneลพeviฤ, Dr Aleksandar Novakovic & Kosta Jovanovic
๐ฆ๐๐ฎ๐ ๐๐๐ป๐ฒ๐ฑ ๐ณ๐ผ๐ฟ ๐บ๐ผ๐ฟ๐ฒ ๐๐ฝ๐ฑ๐ฎ๐๐ฒ๐ ๐ณ๐ฟ๐ผ๐บ ๐๐ถ๐ฟ๐ฐ๐๐ฏ๐ผ๐ ๐ฎ๐ ๐๐ฒ ๐ฎ๐ฝ๐ฝ๐ฟ๐ผ๐ฎ๐ฐ๐ต ๐๐ต๐ฒ ๐ณ๐ถ๐ป๐ฎ๐น ๐ฝ๐ต๐ฎ๐๐ฒ ๐ผ๐ณ ๐๐ต๐ฒ ๐ฝ๐ฟ๐ผ๐ท๐ฒ๐ฐ๐! While the project is nearing its conclusion, this milestone brings exciting new insights, results, and key takeaways that weโll be sharing in the coming period.
This research was supported by the Science Fund of the Republic of Serbia, GREEN โ CircuBot