Nowadays, it is possible to capture the whole surface areas of buildings in three-dimensional point clouds using laser scanners. However, it is still challenging to find the damaged areas of the building data automatically. The objective of the study is to investigate the use of deep neural networks in combination with the building data captured by the terrestrial laser scanner for damage detection. The post-seismic building data for our experiments is from a structural lab. To find out what kinds of DNNs are useful for damage detection, the deep neural networks we choose are PointNet and PointNet++ (MSG, MRG, and SSG). Since innovative DNNs can only directly operate on small 3D rigid models, we transform the detection problem into a classification problem by voxelizing the whole façade before feeding into the 3D DNNs. Meanwhile, we find a pair of optimal parameters after investigating the impact of different point sizes and voxel sizes on detection results. By comparison, we perform a point-based classification by Random Forest to highlight the effectiveness of our approach.

In scallop aquaculture, it is important to monitor the growth of scallops in the sea for growth control. In this paper, we propose a method to automatically detect scallops and determine whether they are alive or dead based on the information obtained by 3D reconstruction of scallops in the underwater from images. The scallops are detected using Sparse ICP, and multiple initial positions are set to deal with the problem of local solutions. The scallops are judged survival by defining an opening angle from the obtained scallop shell positions. The experimental results showed that the scallop detection rate was 100% for the simulation data and 81.3% for the real data. The survival rate was 76.9% for the real data. These results show that the number of samplings is three times larger than that of the previous method, showing the effectiveness of the proposed method.

Peg-in-hole is a task of product assembly. Robots insert the component (called “peg”) into the hole. In this paper, we tackle peg-in-hole with an uncertainty of hole position. Previous works search hole positions to insert pegs into uncertain holes. However, these methods have problems: annotation cost, positional accuracy, and long search time. Therefore, we use imitation learning (IL) which teaches robots through human teleoperation. IL provides the following benefits: lower annotation costs, rough visual servo, and an ability to find accurate hole positions by trial and error of insertion. However, current image-based IL cannot teach force perception to robots without a haptic feedback device. To solve this problem, we focused on pseudo-haptics. Pseudo-haptics is a force presentation method that does not require a mechanical feedback device. We propose a “Pseudo-haptics Module (PHM)” to feel the robot's pseudo-haptics. The module is designed from the process by which humans perceive pseudo-haptics. Our method with PHM is 7.7% higher peg-in-hole success than simple CNN.

This paper improves a deep learning model for point cloud registration of different sizes. Point cloud registration is an important task for applications such as 3D modeling, self-positioning of mobile robots, and environment inspection. Recently, deep learning has started to deal with point clouds. PointNet was the first deep learning model for point cloud classification and semantic segmentation. Since then, methods based on PointNet for tasks such as point clouds registration have also been proposed. However, these methods are only suitable for identical or nearly identical point clouds. The size of the point cloud varies with measurement distance, sensor type, environment, and many other factors. Therefore, point clouds that need to be registered may have very different sizes or dimensions. Conventional methods cannot cope with such situations. In the past, we proposed "PointpartNet", a new deep neural network for point cloud registration based on partial feature extraction. This network specialized in searching the matching region between point clouds of different sizes, followed by a simple registration method. In this paper, we improve this model to increase the accuracy and robustness of point clouds registration. This is done by extracting local point cloud features and passing them through an explicit feature matching network that calculates correspondences between the point clouds to be registered. In qualitative experiments, we demonstrate that this vastly decreases the registration error and increases the success rate of point cloud registration by a s much as 15 to 16 percent, as compared to previous research.

Since the introduction of the Vision Transformer, many transformer-based networks have been proposed. Nakashima et al. showed that ViT and gMLP can be pre-trained in FractalDB and achieve the same level of accuracy as ImageNet-1k. We hypothesize that other Transformer networks may also benefit from pre-training in FractalDB. If this hypothesis is proven, it can be expected that improving FDSL-based datasets such as FractalDB will improve the accuracy of existing networks and those to be proposed in the future. Therefore, in this paper, we perform exhaustive experiments on pre-training results of representative Transformer networks on FractalDB.

Data augmentation is a commonly used method for improving deep learning models in image classification. By adding slightly modified images that do not change the label of the original image to the training data set, the trained model becomes more robust against diverse characteristics of the input image. In this study, we propose a new data augmentation method by improving a previously-known random augmentation method. Our method consists of three steps; 1) determine the set of image modification operators and the number of augmented images, 2) determine the sequence of the image modification operators so that no duplicated sequences are generated, and 3) apply the sequence to augment images. The variety of augmentation is further increased by randomly determining the level (intensity) and the weight of combining the sequences. We applied our method on the CIFAR dataset and show that our method outperforms existing methods.

Diamond is synthesized under the atmospheric pressure and ambient temperature conditions in the present research that aims to construct the in-liquid synthesis technique. The developed technology and the obtained results for an in-liquid synthesis method (iLS method) are as follows; an acetone (organic solvent) dissolves polystyrene, and KOH aqueous solution dissolves and decomposes gelatin/cocoon or rice. The active carbons (included free carbon) and the organic compounds form in an organic sol by chemical change originated between two different liquids solved organic polymers. The organic sol turns its color to black from transparent thin yellow by chemical change between an organic solvent and KOH aqueous solution. The black organic sol appears to contain many free carbons that generated by decomposing active carbons and organic compounds, because an ultraviolet cut their combined bridges in the organic sol. The free carbon precipitated on the diamond substrate forms diamond by the possible adoption of the self-assembly·aggregation. An ultraviolet irradiation leads to precipitate much diamond and less organic polymer and the absence of an ultraviolet causes less diamond and much organic polymer that precipitate on diamond substrate. Diamond is easily to precipitate and aggregate on natural diamond and polycrystalline diamond in mixing organic sol.

This paper discusses the surface modification processing of polished surface of oxygen free copper using NaCl-electrolytic oxidizing water (hereinafter to as EO water). First, the unheated polished surface test pieces were roughened by immersion with and without ultrasonic using NaCl EO water while comparing with the HCl solution. The surface of test pieces after roughening treatment was observed by FE-SEM and measured by AFM. The results showed that NaCl EO water has a stronger roughening effect than HCl solution, and becomes stronger when with ultrasonic. When with ultrasonic, the surface roughness can be controlled by the immersing time into NaCl EO water. Next, the residual abrasive material (Al₂O₃) and chlorine contamination on the surface of the test pieces after treatment were quantitatively evaluated based on EDS analysis results. The results showed that compared with Na₂SO₄ EO water, using NaCl EO water gives the following. Without ultrasonic, the amount of residual abrasive material (Al₂O₃) is similar, and the amount of chlorine contamination is considerably large. With ultrasonic, the amount of residual abrasive material (Al₂O₃) is dramatically reduced, and the amount of chlorine contamination is the same. By this study, we received the suggestion that NaCl EO water could be applied to the surface roughening treatment of polished surface of oxygen free copper.

Light-assisted etching is a processing method that enables selective etching by light illumination onto semiconductors in an etchant. In this study, we propose the cylindrical surface forming technique by light-assisting etching using leaking light from cylindrical glass fibers. When a light propagating glass fiber was placed in contact with a semiconductor wafer, the wafer surface in the vicinity of the contact area was exposed. In addition, by immersing the contacted area in an etchant, selective etching only the exposed area is expected. A glass fiber was contacted with a GaAs wafer and was immersed in nitric acid. Laser with the wavelength of 445 nm entered from the edge face. The profiles of the glass fiber were transferred to the GaAs wafer after the light illumination to the glass fiber. The transferred profile with a curvature radius of ∼50 μm and depth of 0.3∼1.2 μm was formed after 5-minute light exposure. By utilizing various glass fiber with different diameter, cylindrical surface with different curvature radius was formed. The difference in the glass fiber radius and the transferred surface was less than 10%.