Behavioral feature recognition of multi-task compressed sensing with fusion relevance in the Internet of Things environment

Abstract

With the development of electronic and communication technologies, wireless sensors have been widely used. Human behavior recognition based on micro-inertial sensor is an important application of the Internet of Things, and it has received increasing attention. This paper first introduces the theory of compressed sensing and sparse representation to solve the problem of sensor behavior classification. Aiming at the problem of multi-sensor behavior recognition, an effective result fusion method is proposed. By analyzing the multi-task behavior recognition process, the residual model is introduced to effectively integrate the multi-task results and fully exploit the data information. Secondly, in view of the fact that the characteristics of sensor behavior recognition mostly use the time–frequency domain characteristics of digital signal processing, this paper proposes an association feature. In a multi-sensor system, there is a correlation between sensor data at different locations according to human behavior characteristics. The combination of different position sensor information better reflects the human motion characteristics. This feature can effectively mine the potential information in the existing data and improve the behavior recognition rate. Finally, in order to enhance the robustness of the wearable sensing behavior recognition system, the system structure is optimized and analyzed, and the fusion problem of multi-sensor nodes is further studied. In the established decision fusion framework, the adaptive logarithmic optimization pool is used to make decision fusion for the classification posterior probability output of each node, and finally the class of behavior is discriminated. The experimental results show that the proposed method can effectively improve the performance of behavior recognition.