Abstract
Intelligent Transportation Systems (ITS) and autonomous vehicles (AVs) are two fast-moving research fields. Understanding the surroundings and acting quickly is essential for AV control, especially in dynamic environments. Object detection algorithms must function properly under these demanding circumstances. Even though deep learning models are excellent at detecting objects, they are sometimes criticized for having unclear decision-making. Recently, explainable artificial intelligence (XAI) has improved these judgments’ interpretability. This paper proposes a federated learning system that protects data privacy while enabling decentralized training across many AVs of an upgraded object detection algorithm, Yolov7-E. Grad-CAM visualization techniques are used to illustrate the deep learning models, and for easier interpretability, semantic data is fused with object detection. Enhanced security through federated learning, reducing data privacy risks by 40%. Based on comparative studies, it is demonstrated that the suggested model, which utilizes the federated learning paradigm, attains higher mAP and is appropriate for AV decision mechanisms while maintaining data privacy.

This publication has 28 references indexed in Scilit: