Integrating Conventional Models and AI for Intelligent Sensor Data Processing: A Review

Accurate information perception is the foundation for the reliable operation and implementation of intelligent systems and equipment, with its core originating from sensors and data processing technologies. Although the sensor hardware architecture ensures the lower limit of measurement performance, sole reliance on hardware optimization is inadequate for further performance enhancement and functional expansion. Sensor data processing technology is a crucial pathway to advance sensor development, overcoming performance and functional bottlenecks through algorithmic rather than hardware. In recent years, the application of artificial intelligence (AI) tools has significantly elevated the perceptual capabilities of sensor data processing. However, current AI-based sensor data processing research mainly focuses on the data-driven paradigm, over-relying on results directly extracted from sensor data, and neglecting the interpretative induction and prior guidance from conventional model-driven methods. This tendency leads to a black-box process and unnecessary computational resource waste, ignoring the complementary potential between data-driven AI and conventional models. To address these problems, from three perspectives: 1) Single-Sensor Optimization, 2) Cross-Sensor Validation, and 3) Multi-Sensor Cooperative Measurement, this article systematically summarizes the characteristics of AI and model-driven methods for sensor data processing, and further reviews the hybrid research that integrates AI and conventional models to elucidate their synergistic potential. Finally, the future trends and research gaps in sensor data processing are summarized, offering key insights for further studies.