Sensor-based ore sorting can discard barren rock before grinding, but dual-energy X-ray transmission (DE-XRT) grade models are often judged by block-wise error that poorly reflects concentrate and tailings mass balance at the plant cut-off. Process-aware sorting knowledge (PSK) encodes attenuation, sorting tendency, and grade-band text from patch statistics and split rules; captions are aligned via a frozen contrastive language–image pre-training (CLIP) encoder during training, while inference remains Vision-only. Stream outcomes are reported as metal recovery, mass yield (mass fraction to concentrate), concentrate and tailings grades, and enrichment ratio, with the Process Fidelity Index (PFI) summarizing oracle-aligned fidelity at cut-off . On a fixed PbZn holdout with full assay labels, PSK raises the PFI from 0.716 to 0.830 and lowers mean absolute error (MAE) from 4.59% to 3.77% versus Vision-only training; categorical and physics-feature controls support added value from CLIP alignment. Backbone ablation shows positive PSK gains on ResNet-18, MobileNet v3 small, and ShuffleNet v2_×0.5 (PFI = +0.114/+0.118/+0.089), with lightweight Vision PFI near MobileNet and below ResNet-18. Field ablation shows that attenuation and sorting-tendency fields drive most stream gains, while the grade-band field trades a higher PFI for worse tailings grade when removed.
The conventional attenuation-ratio feature R remains affected by particle thickness, material heterogeneity, and beam hardening in practical dual-energy X-ray transmission (DE-XRT) measurements. This study proposes an ore-sorting method based on multi-feature fusion of dual-energy X-ray transmission responses. An aluminium step wedge was used to establish a reference relationship between R and high-energy attenuation . The deviation from this relationship was defined as the residual feature . Particle-level statistics of R, , and were combined in a linear support vector machine (SVM) classifier. For lead–zinc ore, the SVM was compared with a conventional double-threshold R-value rule. In batch-grouped out-of-fold evaluation, the SVM increased combined Pb–Zn recovery from 90.99% to 97.42% and reduced the combined Pb–Zn grade of the tailings from 2.101% to 0.357%. In an independent 216-particle sorting-and-assay test, the SVM recovered 98.78% of Pb–Zn at 22.03% mass rejection, compared with 90.72% recovery at 26.05% rejection for the conventional R-value rule.