Pruning Growing Self-Organizing Map Network for Human Physical Activity Identification.

Human physical activity identification based on wearable sensors is of great significance to human health analysis. A large number of machine learning models have been applied to human physical activity identification and achieved remarkable results. However, most human physical activity identification models can only be trained based on labeled data, and it is difficult to obtain enough labeled data, which leads to weak generalization ability of the model. A Pruning Growing SOM model is proposed in this paper to address the limitations of small-scale labeled dataset, which is unsupervised in the training stage, and then only a small amount of labeled data is used for labeling neurons to reduce dependency on labeled data. In training stage, the inactive neurons in network can be deleted by pruning mechanism, which makes the model more consistent with the data distribution and improves the identification accuracy even on unbalanced dataset, especially for the action categories with poor identification effect. In addition, the pruning mechanism can also speed up the inference of the model by controlling its scale.

Clustering Ensemble Model Based on Self-Organizing Map Network.

This paper proposes a clustering ensemble method that introduces cascade structure into the self-organizing map (SOM) to solve the problem of the poor performance of a single clusterer. Cascaded SOM is an extension of classical SOM combined with the cascaded structure. The method combines the outputs of multiple SOM networks in a cascaded manner using them as an input to another SOM network. It also utilizes the characteristic of high-dimensional data insensitivity to changes in the values of a small number of dimensions to achieve the effect of ignoring part of the SOM network error output. Since the initial parameters of the SOM network and the sample training order are randomly generated, the model does not need to provide different training samples for each SOM network to generate a differentiated SOM clusterer. After testing on several classical datasets, the experimental results show that the model can effectively improve the accuracy of pattern recognition by 4%∼10%.

Divide and conquer: high-accuracy and real-time 3D reconstruction of static objects using multiple-phase-shifted structured light illumination.

Multiple-phase-shifted structured light illumination achieves high-accuracy 3D reconstructions of static objects, while typically it can't achieve real-time phase computation. In this paper, we propose to compute modulations and phases of multiple scans in real time by using divide-and-conquer solutions. First, we categorize total N = KM images into M groups and each group contains K phase equally shifted images; second, we compute the phase of each group; and finally, we obtain the final phase by averaging all the separately computed phases. When K = 3, 4 or 6, we can use integer-valued intensities of images as inputs and build one or M look-up tables storing real-valued phases computed by using arctangent function. Thus, with addition and/or subtraction operations computing indices of the tables, we can directly access the pre-computed phases and avoid time-consuming arctangent computation. Compared with K-step phase measuring profilometry repeated for M times, the proposed is robust to nonlinear distortion of structured light systems. Experiments show that, first, the proposed is of the same accuracy level as the traditional algorithm, and secondly, with employing one core of a central processing unit, compared with the classical 12-step phase measuring profilometry algorithm, for K = 4 and M = 3, the proposed improves phase computation by a factor of 6 ×.