Abstract
Autonomous driving has reached significant milestones in research and development over the last few decades. There is growing attention in the area as the utilization of autonomous vehicles (AVs) assurances safer and more environmentally-friendly transportation systems. Adverse environmental conditions like dust or sandstorms, fog, heavy snow, and rain conditions are unsafe limitations on the function of the cameras by decreasing visibility and influencing driving safety. These limitations affect the performance of tracking and detection methods applied in traffic surveillance systems and autonomous driving applications. Using the fast development in mathematically powerful artificial intelligence (AI) approaches, AVs could sense their atmosphere using higher accuracy, make safer real-world decisions, and work consistently without user intervention. Therefore, this paper proposes a Complex Data Analysis for Adverse Weather Detection in Autonomous Vehicles Using Explainable Artificial Intelligence (CDAAWD-AVXAI) approach. The CDAAWD-AVXAI approach improves the safety and reliability of AVs by ensuring robust weather detection. Initially, the presented CDAAWD-AVXAI approach applies image pre-processing by utilizing the median filter (MF) model to reduce noise and enhance the quality of input images. The CapsNet model is employed for feature extraction to recognize spatial hierarchies and capture intricate patterns from complex visual data. Moreover, the temporal convolutional network (TCN) model detects adverse weather conditions. To further enhance performance, the hyperparameter tuning of the TCN model is performed by implementing the improved dung beetle optimization (IDBO) model. Finally, the XAI using LIME provides transparency and interpretability, allowing stakeholders to understand the decision-making process behind weather condition predictions. The stimulation study of the CDAAWD-AVXAI technique is performed under the DAWN dataset. The experimental validation of the CDAAWD-AVXAI technique portrayed a superior accuracy value of 98.83% over existing methods.