Using the Restricted Boltzmann Machine (RBM) and Deep Belief Network (DBN), a structural adaptive deep learning method was proposed to automatically find the number of hidden neurons and the number of hidden layers considered appropriate for the input data space during learning. Furthermore, a Teacher-Student (T/S) structural adaptive learning method was proposed as an ensemble learning model when there is a mixture of label assignments by multiple annotators, for example, when different teacher signals are assigned to the same input signal. The proposed method has shown higher classification accuracy than conventional methods on various image classification benchmark datasets and has provided an alternative to many time and efforts, such as medical image diagnosis and image recognition in civil engineering and architecture fields. In this paper, we applied this method to the automatic diagnosis of deteriorated areas in the photographs for exterior inspection of power transmission towers using drones. We developed a visualization function based on rectangular detection of abnormal areas and segmentation according to the state of the abnormality. The experimental results showed that rectangle detection was performed with 97.57% accuracy.

As semiconductor design rules progress, the required level of reliability for semiconductor manufacturing equipment is increasing, and it is necessary to find the optimal process conditions that meets the film properties required by customers. The wafers are processed at high temperature close to 800^°C in the deposition process equipment. The experimental design that takes into account interactions between two factors and quadratic term effects is required to find the optimal deposition process conditions, because thermal effects cause these effects. However, since the conventional one factor design experiment cannot reflect these effects on the model. In addition, since the batch deposition process takes nearly 8 hours, it is extremely costly to comprehensively experiment on many factors. In this study, we adopted D-optimal design and definitive screening design as methods of design of experiment for the process conditions optimization of deposition process. Since conducting additional experiments on actual equipment is costly, we built boosting neural network models as virtual semiconductor manufacturing equipment models by using the data from the past experiments. We derived 10 film properties using the Gaussian process regression (GPR) models built by one factor design, D-optimal design and definitive screening design. Then, we compared the differences between the film properties derived by each model and those required by customers. As a result, D-optimal design got the smallest error toward the film properties required by customers, and it was revealed that D-optimal design was suitable for deriving optimal manufacturing process conditions for semiconductor deposition process equipment.

The dynamics of curling stones, the unique feature of curling, need to be fully resolved. In addition, curling tactics still need to be sufficiently analyzed. To conduct these studies, accurate and voluminous stone trajectory data is needed. Although a system combining an IR camera IR-LED and a system using video images has been proposed as a curling stone trajectory measurement system, it is not easy to install the system on an existing curling hall and sheet or to link it with other systems. Therefore, we developed a trajectory measurement system that can be easily installed by estimating its own position and posture. This system can expand the measurement range by combining multiple LRFs that use lasers to detect object positions, as well as estimate the relative positions of the LRFs. The measurement accuracy of the proposed method was determined to be on the order of cm when compared to motion capture.

A new visualization method to help users recognize sleep state misperception is proposed based on hypotheses: (1) that a graphical visualization method can help users become aware of sleep state misperception, and (2) that motivation for continued use can be improved by introducing a qualitative visualization approach. In this method, gaps between subjective and objective evaluations are represented using Mondrian-like abstract graphic patterns. To verify these hypotheses, experiments were conducted and the results were analyzed. The findings support both hypotheses. Hence, the proposed method is comparable to traditional methods and offers an advantage in terms of sustained use.