Waste electrical and electronic equipment (WEEE) represents a heterogeneous waste stream containing valuable non-ferrous metals, yet RGB-based characterization of pre-concentrated output fractions (6–10 mm) recovered from shredded WEEE remains underexplored. Therefore, in this study, we investigate the feasibility of particle-level instance segmentation and propose a process-oriented temporal assessment method. By linking end-to-end detection time (E2E) to particle residence time (PRT) within the sensor’s field of view, the method assesses the preliminary temporal suitability of deep learning (DL) models for future inline applications. For this reason, ten state-of-the-art DL-models (YOLO and Mask R-CNN) were evaluated on a novel particle-level WEEE benchmark dataset. In addition, we examine how visual material characteristics and input variability affect detection performance and the training process. Results demonstrate that YOLOv9 achieved an mAP@0.5 of 75.16% across all material classes, while E2E remained below the PRT, indicating inline feasibility for the investigated application. Replacing the ResNet-50 backbone of Mask R-CNN with VGG-16 reduced E2E from 26 ms to 21.1 ms (53.3%) and increased mAP@0.5 from 67.30% to 74.82%. Our work establishes groundwork for future inline applications, emphasizing the interaction between process constraints and detection time. Overall, these findings advance sensor-based real-time characterization of pre-concentrated fine fractions from shredded WEEE, contributing to future inline applications toward Recycling 4.0.