Root nodule phenotyping is essential for understanding biological nitrogen fixation in legumes, an important process for crop growth and productivity; however, traditional manual approaches for analyzing root nodules can be slow and prone to human bias, making them impractical for large-scale studies. More recent studies have utilized deep learning for high-throughput nodule assessments but usually focus on images of isolated roots. In this study, we implemented and evaluated two state-of-the-art deep learning architectures (YOLOv12 and Mask R-CNN) to automate the detection of root nodules from pea root systems grown within a controlled rhizobox environment. To generate ground truth data, we annotated 62 high-resolution images of root systems, growing within sand-peat media mix. The study also incorporated two complementary Explainable AI (XAI) techniques: Gradient-weighted Class Activation Mapping++ (Grad-CAM++) for class-discriminative explanation and Feature Activation Maps for low-level attention visualization. A comparative evaluation of YOLOv12 and Mask R-CNN architecture shows increased nodule detection performance with YOLOv12 (precision: 59%, recall: 71%) as compared to Mask R-CNN (precision: 46%, recall: 60%), with comparable Intersection over Union (IoU) values (0.61 vs. 0.67). Our implementation of explainable AI techniques through Grad-CAM++ and feature activation visualizations help explain the performance differences observed between the two architectures.