Accurate estimation of wood chip geometric properties is crucial for high-volume biomass manufacturing, where chip size and shape directly impact energy efficiency, product quality, and process optimization. Traditional measurement methods are labor-intensive, inconsistent, and fail to capture chip variability in large-scale industrial settings. This study presents an enhanced DEtection TRansformer (DETR)-based framework for automated wood chip size estimation, offering a scalable, precise, and real-time solution. We have created an extensive dataset comprising 223 images and more than 7,100 annotated wood chips, collected from various sources in the Southeastern United States. The proposed approach consists of (1) deep learning-based object detection using an improved DETR model and (2) size distribution estimation via histograms and statistical metrics. The proposed framework achieves an average mAP50 of 85%, mAR50 of 91%, mAP50-95 of 61%, and recall50-95 of 68% across 5-fold cross-validation on the test set, outperforming existing methods in both accuracy and robustness. This research advances flexible automation and intelligent manufacturing, enabling data-driven process optimization for industrial biomass applications.