The increasing deployment of Ultra-High Frequency (UHF) RFID systems in applications such as logistics, retail, and industrial automation has heightened the need for efficient, real-time data cleansing, particularly on embedded middleware platforms. One of the most persistent challenges in such systems is temporal data redundancy, where identical RFID tags are read repeatedly within short intervals. This results in increased memory consumption, processing delays, and potential data integrity issues, especially on embedded devices with limited computational resources. While existing deduplication methods, such as Bloom filters and sliding windows, have shown promise in server-based environments, they are not fully optimized for edge computing scenarios due to their resource overhead and risk of false positives.
To address this gap, this research proposes a lightweight, embedded-oriented RFID middleware system based on a modified xxHash32 hashing algorithm and an idle-time-based sliding window mechanism. The deduplication process is triggered only during active RFID data streams, significantly reducing memory usage and processing overhead. The system also incorporates real-time NTP-based time synchronization, precise timestamp labeling, efficient serial data parsing, and structured JSON output formatting for downstream processing. Experimental evaluations were conducted across 288 test scenarios involving 12 datasets with varying levels of redundancy and two data rates (100 and 1100 tags per second). Results show that the proposed modified xxHash32 algorithm outperforms 11 baseline deduplication methods, achieving 100% accuracy, the lowest average latency (14.2 ?s), and the highest throughput (up to 883,555 tags/sec) while consuming minimal memory and flash storage. These findings validate the proposed system's scalability, responsiveness, and suitability for resource-constrained embedded RFID applications, providing clean, synchronized data for integration with IoT or Complex Event Processing (CEP) platforms