Abstract
<jats:sec> <jats:title>Abstract</jats:title> <jats:p> As the British steel industry transitions to a scrap-based electric arc furnace route for steelmaking, steel scrap quality is becoming increasingly important. The chemical composition of fragmentized scrap, particularly the weight percentage of elements such as copper, strongly governs its suitability for high-quality steelmaking. However, existing chemical characterization techniques are too expensive and operationally impractical to deploy across the entire scrap stream, leaving composition estimates vulnerable to fluctuations in tramp element distributions introduced by unwanted residuals. To enable fast, cost-effective assessment, this work uses copper as an exemplar and investigates image-based classification for tramp element composition estimation for fragmentized scrap. Using commercial-grade scrap, an image dataset was manually curated to capture diverse mixtures of steel and copper fragments, with corresponding weight measurements and segmentation annotations collected for each sample image. A neural network was then trained using image segmentation, with the predicted segmentation maps provided as input to a classifier, achieving <jats:inline-formula> <jats:alternatives> <jats:tex-math>$$86.67\%$$</jats:tex-math> <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mn>86.67</mml:mn> <mml:mo>%</mml:mo> </mml:mrow> </mml:math> </jats:alternatives> </jats:inline-formula> accuracy in assigning images to the correct copper composition class and demonstrating the feasibility of computer vision methods for tramp element composition analysis in scrap quality monitoring. </jats:p> </jats:sec>