List of Publications
Compound-protein Interaction Prediction with End-to-end Learning of Neural Networks for Graphs and Sequences
Yu Oishi, Hiroyuki Oguma, Ayako Tamura, Ryosuke Nakamura and Tsuneo Matsunaga
Information about changes in the population sizes of wild animals is extremely important for conservation and management. Wild animal populations have been estimated using statistical methods, but it is difficult to apply such methods to large areas. To address this problem, we have developed several support systems for the automated detection of wild animals in remote sensing images. In this study, we applied one of the developed algorithms, the computer-aided detection of moving wild animals (DWA) algorithm, to thermal remote sensing images. We also performed several analyses to confirm that the DWA algorithm is useful for thermal images and to clarify the optimal conditions for obtaining thermal images (during predawn hours and on overcast days). We developed a method based on the algorithm to extract moving wild animals from thermal remote sensing images. Then, accuracy was evaluated by applying the method to airborne thermal images in a wide area. We found that the producer's accuracy of the method was approximately 77.3% and the user's accuracy of the method was approximately 29.3%. This means that the proposed method can reduce the person-hours required to survey moving wild animals from large numbers of thermal remote sensing images. Furthermore, we confirmed the extracted sika deer candidates in a pair of images and found 24 moving objects that were not identified by visual inspection by an expert. Therefore, the proposed method can also reduce oversight when identifying moving wild animals. The detection accuracy is expected to increase by setting suitable observation conditions for surveying moving wild animals. Accordingly, we also discuss the required observation conditions. The discussions about the required observation conditions would be extremely useful for people monitoring animal population changes using thermal remote sensing images.
Takashi Ito, Masateru Ishiguro, Tomoko Arai, Masataka Imai, Tomohiko Sekiguchi, Yoonsoo P. Bach, Yuna G. Kwon, Masanori Kobayashi, Ryo Ishimaru, Hiroyuki Naito, Makoto Watanabe & Kiyoshi Kuramoto
The near-Earth asteroid (3200) Phaethon is the parent body of the Geminid meteor stream. Phaethon is also an active asteroid with a very blue spectrum. We conducted polarimetric observations of this asteroid over a wide range of solar phase angles α during its close approach to the Earth in autumn 2016. Our observation revealed that Phaethon exhibits extremely large linear polarization: P = 50.0 ± 1.1% at α = 106.5°, and its maximum is even larger. The strong polarization implies that Phaethon's geometric albedo is lower than the current estimate obtained through radiometric observation. This possibility stems from the potential uncertainty in Phaethon's absolute magnitude. An alternative possibility is that relatively large grains (~300 μm in diameter, presumably due to extensive heating near its perihelion) dominate this asteroid's surface. In addition, the asteroid's surface porosity, if it is substantially large, can also be an effective cause of this polarization.
H. Zhang, X. Song, T. Xia, M. Yuan, Z. Fan, R. Shibasaki, Y. Liang, "Battery electric vehicles in Japan: Human mobile behavior based adoption potential analysis and policy target response", Applied Energy, Volume 220, Pages 527- 535
Nevrez Imamoglu , Yu Oishi, Xiaoqiang Zhang , Guanqun Ding, Yuming Fang, Toru Kouyama , Ryousuke Nakamura
<Abstract> Many works have been done on salient object detection using supervised or unsupervised approaches on color images. Recently, a few studies demonstrated that efficient salient object detection can also be implemented by using spectral features in visible spectrum of hyperspectral images from natural scenes. However, these models on hyperspectral salient object detection were tested with a very few number of data selected from various online public dataset, which are not specifically created for object detection purposes. Therefore, here, we aim to contribute to the field by releasing a hyperspectral salient object detection dataset with a collection of 60 hyperspectral images with their respective ground-truth binary images and representative rendered color images (sRGB). We took several aspects in consideration during the data collection such as variation in object size, number of objects, foreground-background contrast, object position on the image, and etc. Then, we prepared ground truth binary images for each hyperspectral data, where salient objects are labeled on the images. Finally, we did performance evaluation using Area Under Curve (AUC) metric on some existing hyperspectral saliency detection models in literature.
Localization of Dark Mantle Deposit on the Lunar Surface with Shadow Masked Multiband Imager Data
Riho ITO, Hiroka INOUE, Makiko OHTAKE, Yoshiaki ISHIHARA, Hisashi OTAKE, Ryosuke NAKAMURA
Global survey of the low-reflectance region, known as dark mantle deposits (DMD), is an important challenge for lunar geological science. While smaller DMDs can be detected by using high-resolution lunar images, the automatic identification would be hindered by shadows. The purpose of this study is a full-scale search of DMDs by using Kaguya's Multiband Imager (MI) data with complete shadow masks generated by deep learning.
Koichi TSURU, Riho ITO, Chikatoshi HONDA, Ryosuke NAKAMURA
In lunar geology, it is one of the most basic and important tasks to detect lunar surface features (e.g., craters, boulders, ridges, etc.) from huge observation data brought by lunar explorers such as SELENE or Lunar Reconnaissance Orbiter. So far, various methods for automatic detection from the images and digital elevation model (DEM) have been proposed. Among them, deep learning shows much higher performance than other methods. The biggest shortcoming of deep learning, however, is the huge manual work required to create a large-scale training data set. In this study, we investigated the methods for automatic detection of craters from satellite imagery by using two models based on Generative Adversarial Networks (GANs).
Renewable energy has been actively introduced around the world in recent years. In particular, the number of photovoltaic power plants will steadily increase because of the lower construction and operation cost than other power plants. According to the International Energy Agency's report, the total electricity generated by photovoltaic power plants, which were newly constructed in 2016, amounted to 75 GW. The global survey based on satellite imagery is useful for a variety of applications, such as design of an efficient transmission network, accurate estimate of the total power generation, detection of newly established or abolished photovoltaic power plant, and so on.
The Greenhouse Gases Observing Satellite (GOSAT) was launched in 2009 to measure global atmospheric CO2 and CH4concentrations. GOSAT is equipped with two sensors: the Thermal And Near infrared Sensor for carbon Observations (TANSO)-Fourier transform spectrometer (FTS) and TANSO-Cloud and Aerosol Imager (CAI). The presence of clouds in the instantaneous field of view of the FTS leads to incorrect estimates of the concentrations. Thus, the FTS data suspected to have cloud contamination must be identified by a CAI cloud discrimination algorithm and rejected. Conversely, overestimating clouds reduces the amount of FTS data that can be used to estimate greenhouse gas concentrations. This is a serious problem in tropical rainforest regions, such as the Amazon, where the amount of useable FTS data is small because of cloud cover. Preparations are continuing for the launch of the GOSAT-2 in fiscal year 2018. To improve the accuracy of the estimates of greenhouse gases concentrations, we need to refine the existing CAI cloud discrimination algorithm: Cloud and Aerosol Unbiased Decision Intellectual Algorithm (CLAUDIA1). A new cloud discrimination algorithm using a support vector machine (CLAUDIA3) was developed and presented in another paper. Although the use of visual inspection of clouds as a standard for judging is not practical for screening a full satellite data set, it has the advantage of allowing for locally optimized thresholds, while CLAUDIA1 and -3 use common global thresholds. Thus, the accuracy of visual inspection is better than that of these algorithms in most regions, with the exception of snow- and ice-covered surfaces, where there is not enough spectral contrast to identify cloud. In other words, visual inspection results can be used as truth data for accuracy evaluation of CLAUDIA1 and -3. For this reason visual inspection can be used for the truth metric for the cloud discrimination verification exercise. In this study, we compared CLAUDIA1-CAI and CLAUDIA3-CAI for various land cover types, and evaluated the accuracy of CLAUDIA3-CAI by comparing both CLAUDIA1-CAI and CLAUDIA3-CAI with visual inspection (400 × 400 pixels) of the same CAI images in tropical rainforests. Comparative results between CLAUDIA1-CAI and CLAUDIA3-CAI for various land cover types indicated that CLAUDIA3-CAI had a tendency to identify bright surface and optically thin clouds. However, CLAUDIA3-CAI had a tendency to misjudge the edges of clouds compared with CLAUDIA1-CAI. The accuracy of CLAUDIA3-CAI was approximately 89.5 % in tropical rainforests, which is greater than that of CLAUDIA1-CAI (85.9 %) for the test cases presented here.
Yu Oishi, Yoshito Sawada, Akihide Kamei, Kazutaka Murakami, Ryosuke Nakamura and Tsuneo Matsunaga
Greenhouse Gases Observing SATellite-2 (GOSAT-2) will be launched in fiscal year 2018. GOSAT-2 will be equipped with two Earth-observing instruments: the Thermal and Near-infrared Sensor for carbon Observation Fourier Transform Spectrometer 2 (TANSO-FTS-2) and TANSO-Cloud and Aerosol Imager 2 (CAI-2). CAI-2 can be used to perform cloud discrimination in each band. The cloud discrimination algorithm uses minimum reflectance (Rmin) for comparisons with observed top-of-atmosphere reflectance. The creation of cloud-free Rmin requires 10 CAI or CAI-2 data. Thus, Rmin is created from CAI L1B data for a 30-day period in GOSAT, with a revisit time of 3 days. It is necessary to change the way in which 10 observations are chosen for GOSAT-2, which has a revisit time of 6 days. Additionally, Rmin processing for GOSAT CAI data was updated to version 02.00 in December 2016. Along with this change, the resolution of Rmin changed from 1/30◦ to 500 m. We examined the impact of changes in Rmin on cloud discrimination results using GOSAT CAI data. In particular, we performed comparisons of: (1) Rmin calculated using different methods to choose the 10 observations and (2) Rmin calculated using different generation procedures and spatial resolutions. The results were as follows: (1) The impact of using different methods to choose the 10 observations on cloud discrimination results was small, except for a few cases, e.g., snow-covered regions and sun-glint regions; (2) Cloud discrimination results using Rmin in version 02.00 were better than results obtained using Rmin in the previous version, apart from some special situations. The main causes of this were as follows: (1) The change of used band from band 2 to band 1 for Rmin calculation; (2) The change of spatial resolution of Rmin from 1/30° to 500-m.