Human-computer interaction based interface design of intelligent health detection using PCANet and multi-sensor information fusion

Abstract

Background and objective: At present, because health monitoring using human-computer interaction (HCI) has become a demand in society, an intelligent health detector with HCI characteristics is urgently needed. Our device and software framework can provide natural human-machine interaction and facilitate the use of effective, efficient and safe electronic equipment for health detection. This paper integrates the research results of human subjects testing with analysis of our computational algorithms to build the proposed interaction platform.

Methods: We collected the pulse signals of normal and sub-health people and used them as pre-processed signals. Then, we input them into the Principal Componcent Analysis Network (PCANet) layer by layer, and extracted the corresponding mapping features in each layer. The extracted features are hash coded, and histogram blocks are used as the feature matrix. Next, the accuracy obtained by the classical classifier is compared with the classification results of other feature extraction methods. The HCI integrated intelligent health detector based on PCANet neural network and multi-sensor information fusion has significantly improved the accuracy of human health detection.

Results: The experimental results show that the proposed method achieves high accuracy for sub-health state recognition. Compared with the traditional feature extraction method, our PCANet method improves the recognition rate by more than 10%, which proves the effectiveness of PCANet model in the field of sub-health pulse signal detection. Because the PCANet is a multi-layer architecture model, in order to verify the influence of the number of extended network layers on the experimental results, experiments are carried out on the three-layer architectures represented by PCANet-1, PCANet-2 and PCANet-3 respectively. The Experimental results show that PCANet-3 model is 2.4% higher than PCANet-1, but only 0.6% higher than PCANet-2. The running time is about 2 times and 1.6 times higher than that of PCANet-1 and PCANet-2; Compared with traditional feature extraction algorithms such as MP and Gabor transform, the accuracy of pcanet model in pulse signal sub-health detection is significantly improved. Therefore, this product can effectively distinguish between the health and sub-health states.

Conclusion: Our research shows that the intelligent health detector is efficient and convenient to use, and has higher accuracy for health detection. The HCI integrated platform provides a new reference basis for the detection of sub-health state.

Keywords: Deep learning; Health detector; Human-computer interaction; Interface design; PCANet.