ABSTRACT The rapid development of smart cities has amplified the need for intelligent parking management systems capable of accurately detecting parking slots and classifying their occupancy status in real time. However, conventional centralized learning approaches often compromise privacy and scalability, as raw sensor and image data must be transmitted to cloud servers. To overcome these limitations, this study proposes a new MSFTNet‐based federated privacy‐preserving framework for smart parking slot detection and occupancy classification. The proposed Multi‐Scale Fusion Transformer Network (MSFTNet) combines convolutional and Transformer‐based feature encoders to capture both fine‐grained spatial textures and global contextual representations, ensuring robust detection under varying lighting, occlusion, and weather conditions. The framework is trained in a federated learning environment, enabling decentralized model updates across multiple edge devices without exposing raw data, while a differential privacy layer and secure aggregation protocol further enhance data confidentiality during parameter exchange. The model was evaluated on the CNRPark‐EXT and PKLot datasets, two widely recognized benchmarks for parking slot analysis containing over 50,000 labeled images captured under diverse real‐world conditions. Experimental results demonstrate that the proposed framework achieves 97.3% accuracy and an F1‐score of 95.8% for slot occupancy classification, outperforming existing CNN and Transformer‐based baselines. Moreover, the federated configuration reduced communication overhead by 22% while effectively preventing privacy leakage. The multi‐scale fusion strategy results in an accuracy gain of 2.1% and an F1‐score gain of 2.4%, confirming its critical role in improving feature representation. These findings validate that the MSFTNet‐based federated architecture provides an optimal balance between performance, privacy, and scalability, making it a highly effective solution for next‐generation smart parking systems within intelligent transportation infrastructures.

Wireless Sensor Networks (WSNs) are increasingly deployed in diverse commercial and industrial IoT applications, yet they remain vulnerable to security threats due to their wireless nature and resource constraints. While Information-Centric Networking (ICN) architectures improve data-centric security over conventional IP networks, internal attacks continue to challenge network reliability. To address these issues, we propose EPSTM, an Evolutionary Particle Swarm Optimized Trust Management Scheme, for authenticating IoT nodes and mitigating malicious behavior. The scheme evaluates trust using network-specific metrics, including device proximity, energy consumption, data transmission reliability, and message delivery timing, to guide secure routing decisions. Simulation results demonstrate that the proposed framework outperforms current state-of-the-art techniques, achieving reduced response times, minimized authentication delays, and lower request volumes from compromised nodes. Notably, for a network of 100 nodes, EPSTM achieved an accuracy of 98.66%, highlighting its efficacy in enabling secure, energy-efficient, and trustworthy IoT communication.