Yamanaka Takao

In recent years, the deep learning techniques have been applied to the estimation of saliency maps, which represent probability density functions of fixations when people look at the images. Although the methods of saliency-map estimation have been actively studied for 2-dimensional planer images, the methods for omni-directional images to be utilized in virtual environments had not been studied, until a competition of saliency-map estimation for the omni-directional images was held in ICME2017. In this paper, novel methods for estimating saliency maps for the omni-directional images are proposed considering the properties of prior distributions for fixations in the planar images and the omni-directional images.

In this paper, we propose a fully convolutional DenseNet model for saliency-map prediction (DenseSal). While the most state-of-The-Art models for predicting saliency maps use shallow networks such as VGG-16, our model uses densely connected convolutional networks (DenseNet) with over 150 layers. Since DenseNet has shown the excellent results on image classification tasks, the coarse feature maps from the fully convolutional neural networks based on DenseNets were concatenated to predict saliency maps through a readout network. It is shown that the DenseNet is useful for the saliency-map prediction and achieved the state-of-The-Art accuracy on the major fixation datasets.

Gas sensors have been widely used for various applications, such as gas leak detection, fire alarm systems, and odor-sensing systems. A problem of the gas sensors has been the selectivity to a target gas: background gases interfere with the measurement of the target gas. In the human olfaction, sensitivity to background odors is decreased by adaptation to the odors. Recently, several bio-inspired signal-processing methods mimicking the adaptation mechanism have been proposed for improving the selectivity of the gas sensors. In this chapter, the studies on the bio-inspired background suppression methods are reviewed. Furthermore, a case study of the bio-inspired background suppression is introduced. In the case study, a perceptron neural network with anti-Hebbian learning was used for realizing the adaptation to the background gas, and was implemented into a digital circuit for real-time gas sensing.

The authors of this chapter study the odor reproduction system synchronously with a movie. The system is made up of an odor sensing system and an olfactory display. The fruit flavors were recorded with movies using a digital video camera and the odor sensing system. The results of the sensory tests showed that the odor information recorded using the proposed method is appropriate for the smell regeneration associated with the movie. Next, the authors propose a tele-olfaction system synchronous with visual information. The olfactory display system was placed remotely from the odor sensing system, and both of them were connected via Internet. In addition to the olfactory system, a Web camera captures image around the sniffing point and that image appears at the computer display connected to the olfactory display at remote site. Moreover, the mobile stage with its sniffing point and the Web camera remotely controlled by a user was introduced so that he/she could interactively approach a smelling object. The questionnaire survey at the exhibition revealed that a user can enjoy smell synchronous with movie in real time even if he/she stays at the remote site. © 2011, IGI Global.

In this study, a new method for suppressing the background odor effect is proposed. Since odor sensors response to background odors in addition to a target odor, it is difficult to detect the target odor information. In the conventional odor sensing systems, the effect of the background odors are compensated by subtracting the response to the background odors (the baseline response). Although this simple subtraction method is effective for constant background odors, it fails in the compensation for time-varying background odors. The proposed method for the background suppression is effective even for the time-varying background odors. © 2008 The Institute of Electrical Engineers of Japan.

In this paper, a method of concurrent recording and regeneration of visual and olfactory information is presented using electronic nose technology. To accomplish this objective, the sensor response patterns of odors in the atmosphere were measured using QCM (quartz crystal microbalance) odor sensors with partially overlapping specificities. Then the odors were identified from the response patterns using LVQ (learning vector quantization), a pattern classification algorithm of neural networks with supervised learning. Visual information, presented as a movie, was captured using a digital video camera; concurrently, odors using odor sensor responses were paired to the video. The recorded visual and olfactory information was evaluated by sensory tests to investigate the effectiveness of the proposed system. As a result, it was found that the olfactory information recorded using the proposed method was appropriate for odor regeneration associated with the movie.

We propose a biologically inspired signal processing model capable of enhancing the discrimination of multivariate patterns from gas sensor arrays. The model captures two functions in the early olfactory pathway: chemotopic convergence of sensory neurons onto the olfactory bulb, and center on-off surround lateral interactions. Sensor features are first topologically projected onto a two-dimensional lattice according to their selectivity profile, leading to odor-specific spatial patterning. The resulting patterns serve as inputs to a network of mitral cells with center on-off surround lateral inhibition, which enhances the initial contrast among odors and decouples odor identity from intensity. The model is validated using experimental data from an array of temperature-modulated metal-oxide sensors. Our results indicate that the model is able to improve the separability between odor patterns that is available at the inputs. (c) 2006 Elsevier B.V. All rights reserved.

In this paper. we propose a new technique for feature extraction/selection based on the projection of sensor features in class space while taking into account the sensor variance. The proposed technique is inspired by the organization of the early stages in the biological olfactory system. The algorithm proves to be highly suitable for high-dimensional feature vectors. The performance shows robustness with problems where only a small number of samples are available as a training dataset. We demonstrate the method on experimental data from two metal oxide sensors driven by a sinusoidal temperature profile. (c) 2006 Elsevier B.V. All rights reserved.

Despite high significance of the sense of smell, the technique of recording and reproducing the olfactory modality was hardly investigated. The odor recorder developed by our group enables electronic recording of the odors and their reproduction at any time and at properly equipped place. The odor recording technique is expected to be useful in many fields such as virtual environments and e-commerce. To realize the future application of the odor recorder, the range of flavors to be handled by the instrument has to be significantly expanded. However, the expansion has to meet the constraint of reasonable number of odor components. One way to achieve that is using the approximation technique with significant reduction of the flavor components. The alternative way could be blending the complex mixtures of odorants treated as single components. In the present study, we focused on investigating both approaches. In the first part of experiment, we successfully recorded the typical orange flavor, approximating the smell with only 3 of 14 components. Subsequently performed human sensory tests (triangle test) confirmed similarity of reproduced flavor and the target one. In the second part of the experiment, three binary mixtures of essential oils extracted from citrus fruits were successfully recorded using the odor recorder. Results of both parts of the study allowed further expansion of the applicable range of flavors for odor recorder. © 2004 Elsevier B.V. All rights reserved.

A smell reproduction technique is useful in the field of virtual reality. We have developed the system called an odor recorder for reproducing the smell recorded using the odor sensing technique. We proposed the new type of the odor recorder using the inkjet devices together with a mesh heater. Droplets with tiny volume were forcibly evaporated to generate smell rapidly and reproducibly. Moreover, the mesh heater was directly connected to the sensors without plumbing tubes and the sensors were placed away from the wall of sensor cell. The recording time of the odor with high odor intensity became much shorter than that of the previous system. Then, the recipe of jasmine scent composed of benzyl acetate and Ylang Ylang was successfully determined using the proposed system.

This paper presents the electronic record of dynamically changing odors under environmental disturbance using a new recording method for the odor recorder called 'real-time reference method'. The target odor is recorded as the recipe of several odor components. Once the recipe to represent the target odor is determined, it can be reproduced anytime and anywhere. In the real-time reference method, the target and blended odors are alternately exposed to the sensor array every sampling interval to determine the recipe for reproducing the target odor. Since the two odors are measured in the almost same environmental condition, the influence of dynamically changing environmental disturbance (temperature and humidity disturbances) can be reduced. Moreover, the dynamically changing target odor can be recorded because of the real-time monitoring of the target odors. In this paper, the dynamically changing apple flavors composed of four components were successfully quantified under the temperature and humidity disturbances. (C) 2003 Elsevier Science B.V. All rights reserved.

A new method for the odor recorder, electronically recording the recipe. of odors or scents made up of many components, was proposed. Although the apple flavors have been recorded with the mixture of five components using the odor recorder in previous work, the number of odor components should be increased to expand the applicable range of odors. Since the collinearity problem of the odor sensor array became apparent with the increase of the components, the new method based on the variable transformation using singular value decomposition was developed in this paper to extract the effective subspace of the sensor outputs for the recipe exploration. As a result, the sensor-array response pattern of the reproduced odor with the eight-component recipe almost agreed with that of the target apple flavor. Furthermore, the human sensory tests revealed that the smell of the approximated odor was identical to that of the target flavor.

The odor recorder has been developed using the odor sensing system based on the active sensing method to electronically record odors and reproduce them anytime and anywhere. The recorder developed in this paper has capability of recording the target odor, the mixture of five odor components, using the odor sensor array composed of quartz crystal microbalance sensors. The difference between the sensor response patterns of the recipe and target odors was minimized so that the recipe of the five components for the target odor was determined. In this paper, the method of selecting the most appropriate sensors to record odors was investigated based on the condition number of the sensor array sensitivity matrix. Moreover, the method of determining the parameters of the odor-recording algorithm was studied. As a result, several apple flavors composed of the five components were successfully recorded. Furthermore, the typical apple flavor composed of nine components was approximated using the mixture of the five components. The approximation result by the sensing system was successfully verified by the human sensory test. (C) 2002 Elsevier Science B.V. All rights reserved.

A new algorithm to electronically record an odor composed of the mixture of many components using the odor sensing system was proposed. Since the human impression on the odor mostly depends on the kinds of key components rather than the accurate mixture composition, the odor was approximated with the combination of several components. Moreover, the efficient method to explore the recipe for the odor approximation with the combination of many components was established. Then, the recipe of an apple flavor was recorded using the proposed method and was evaluated by the human sensory test. As a result, it was found that the odor approximated with up to eight components was close to the target apple flavor. Furthermore, it was confirmed that the response pattern of the QCM (quartz crystal microbalance) odor sensor array was a good indicator of the odors detected by human olfaction. The proposed method was useful to realize the odor recorder with many odor components. (C) 2002 Elsevier Science B.V. All rights reserved.

An odor recorder has the capability of not only recording odors electronically but also reproducing it anytime and anywhere. In this paper, we developed a new type of odor recorder, composed of the odor blenders that generate the mixture of components at any composition using solenoid valves controlled by the delta-sigma modulation method. We optimized the parameters of modulation and the characteristic of the low pass filter by simulation. Moreover, the recipes of apple flavor, jasmine scent, and spice flavor were successfully determined using the developed odor recorder by experiment. The new method is suitable for realizing an odor recorder for many components. (C) 2002 Elsevier Science B.V. All rights reserved.

A new method to find a gas/odor source is proposed. A portable homogeneous gas sensor array is used to visualize the flow of a target gas. The direction of a gas source is estimated using a real-time image processing algorithm, and the source is located by following the estimated direction. The design of the sensor array has been performed using the simulation technique developed in the previous study. In the designing process, a virtual sensor array with an arbitrary time constant is placed in a dynamic gas distribution field visualized using a tracer such as smoke of joss sticks. The virtual sensor array behavior, which is much slower than the gas concentration change, is then calculated by inputting the optical data, captured by a video camera, to the sensor dynamics model. As a result, the designed system has been found effective through the simulation. (C) 2000 Elsevier Science S.A. All rights reserved.

A new method to find a gas/odor source is proposed. A portable gas sensor array was fabricated by arranging 5 × 5 pulse drive semiconductor gas sensors, and was used to visualize the flow of a target gas. The direction of the flow is estimated by an image processing algorithm, and the gas source can be located by tracking the obtained direction. The redundant information from the sensor array is expected to enhance the reliability of the obtained direction compared with the case of a few sensors, especially under the environment with large wind turbulence and/or a quite slow wind field. The experiment in a wind tunnel revealed the soundness of the basic idea. The direction of the gas flow of 1 ~ 2cm/s, lower than the limit of our previous system, has successfully been estimated by the fabricated system. © 1999, The Institute of Electrical Engineers of Japan. All rights reserved.

A new method to analyze the transient response of a gas sensor is proposed. For the analysis of the sensor dynamics, the actual rapid gas concentration change during the measurement of sensor transient response should be obtained. Therefore, the distribution of the gas concentration around the sensor is visualized using the white smoke of joss sticks, and is measured as the brightness of the CCD camera image. The measured sensor response is found to be well described by a model with two phases, such as the response and recovery phases, and their transients in the two phases are expressed as second order linear systems. The results show the feasibility of estimating the sensor response from the visualized gas distribution image. (C) 1998 Elsevier Science S.A. All rights reserved.

A new method to find a gas/odor source is proposed. A portable homogeneous gas sensor array is used to visualize the flow of a target gas. The direction of a gas source is estimated using a real-time image processing algorithm, and the source is located by following the estimated direction. The design of the sensor array has been performed using the simulation technique developed in the previous study. In the designing process,a virtual sensor array with an arbitrary time constant is placed in a dynamic gas distribution field visualized using a tracer such as smoke of joss sticks. The virtual sensor array behavior much slower than the gas concentration change is then calculated by inputting the optical data, captured by a video camera, to the sensor dynamics model. As a result, the designed system has been found effective through the simulation.