To inspect prescription drugs with press-through package (PTP), we propose an automated inspection system which based on computer vision. In the proposed system, we capture PTP drugs and apply hierarchical identification consist of several weak classifiers. In this paper, we report several results of inspection experiments which distinguish about a thousand kinds of PTPs. As a result, we have achieved sufficient recognition rate and processing time.
Cell analysis is an important technology that is widely used for medical diagnosis in hospitals and cell engineering research. Among cell analysis technology, dielectrophoresis (DEP) is one of the most promising approaches for discriminating biological particles because this phenomena requires no labeling procedure with a fluorescent dye or magnetic beads. In this study, we developed a precise cell analysis system by evaluating the DEP force with a liquid flow system. The DEP forces acting on a cell was characterized using a microfluidic chamber containing an electrode-array and fluid-induced shear forces. On the basis of this characterization, the discrimination between healthy skin cells and cancer cells was performed using our novel DEP cell sorting system. As a result, the healthy cells and cancer cells could be distinguished even if the dielectrophoretic properties of those cells were similar.
Victory or defeat in team sports depends on each player's technique, physical strength, and psychological condition. It follows that team performance depends on the player's adaptation to (suitability for) a certain role (position in the team) and the relationships between different roles. We assume that team performance is related to physical and psychological features. Many researchers have proposed that physical features determine a player's suitability for a position. Psychological features have also been researched as factors of position adaptation. However, each feature has been investigated independently. The present research aims to develop a clustering method that considers both physical and psychological features in judging an individual's role and adaptation in the game. This paper reports the concept of the algorithm and result of psychological data analysis using self-organizing maps and principal component analysis.
This paper proposed override ship maneuvering simulator using actual training ship for young pilot trainees. In this new simulator, augmented reality toolkit was used to reproduce scenery from the bridge of large vessel on the actual training ship. The effectiveness of the developed control system which reproduced the large vessel's maneuverability under wind disturbance with an actual training was indicated with results of various simulation experiments.
We have developed the tactile sensing system including haptic bidirectionality for laparoscopic surgery. In humans' active touch, exploration influences perception, but also perception influences exploration. Humans can optimize or change the exploratory movements according to the perception and/or the task, consciously or unconsciously. Our proposed sensing system uses the user's ability on haptic perception by including the bidirectionality. In this paper, considering the actual situation in the application to the laparoscopic surgery, a slim forceps-shaped sensor is assembled and a simulated stomach wall with the lump is used for experiments. The sensor output for the lump is discussed and a signal processing of the collected sensor output for the lump detection is presented. Results show a potential of the tactile sensing system.
This paper presents the Average-Max Reinforcement Learning (AMRL) algorithm that can be used to approximate a global policy of a Markov Decision Process (MDP) as a set of local policies that can be executed in a partially observable environment. The local policies are obtained by reinforcement learning and averaging state-action tables under a stochastic process model. This approach overcomes the scalability problem that arises when a large MDP has to be solved exactly. The approach is motivated by the problem of computing coordination policies for correlated but distributed sensors. We demonstrate the performance of this learning scheme on a simulation of a wireless body sensor network. These results show that the performance of the AMRL algorithm is significantly better than a random policy and is close to the optimal policy that can be obtained from solving a global MDP. The results also show that the AMRL algorithm is scalable to networks represented by large state spaces.
In automation systems and sensor networks, powering the sensor nodes autonomously and transmitting information is a challenge. Effectively configuring supercapacitor banks at each sensor node with suitable energy harvesting systems can assist in such situations, to avoid the longer-term issues with batteries. With modern supercapacitor families with extra low equivalent series resistance (ESR) and energy storage capacities in the range of few joules to several kilojoules could be used effectively, if the electronics and processing requirements at the sensor node is designed with low-power approaches. The paper presents an overview of present day supercapacitor capabilities, and approaches to design low-power-low-energy electronics where sensor nodes could be autonomous in operation. Few examples of energy harvesters coupled with supercapacitors will be presented with some experimental data collected for a proof-of-concept project.
The proposed novel Genetic Bees Algorithm (GBA) is an enhancement to the swarm-based Bees Algorithm (BA). It is called the Genetic Bees Algorithm because it has genetic operators. The structure of the GBA compared to the basic BA has two extra components namely, a Reinforced Global Search and a Jumping Function. The main advantage of adding the genetic operators to BA is that it will help the algorithm to avoid getting stuck in local optima. In this study the scheduling problem of a single machine was considered. When the basic BA was applied to solve this problem its performance was affected by its weakness in conducting global search to explore the search space. However, in most cases the proposed GBA overcame this issue due to the two new components which have been introduced.
This paper proposes a method to identify the spatial distribution patterns of cone cells related with blood vessel in a given retina image. We define three types of the distribution patterns between cones and vessels. Positive correlation distribution (PCD) and negative correlation distribution (NCD) indicate that the cones tend to be close to or far from the vessels. While the cone cells do not have significant correlation with vessels, the cone distribution is regarded as the random distribution (RD). In our method, RD is modeled by many virtual retina images, each of which is generated by the vessels extracted from the original retina image and the virtual cells are selected randomly from the image. Using the virtual images, we estimate the distribution range of RD. When the distribution of the original cells is above the upper limit or below the lower limit of the RD distribution, the cell distribution is NCD or PCD. Otherwise, the cell distribution is regarded as RD.
In many applications of neural networks, e.g. time series prediction or pattern analysis, training data are generated automatically out of large data sets. The problem is to determine the varying significance of the resulting training vectors concerning the given task in order to make appropriate decisions for the training phase. In this paper we propose a self-organized significance analysis based on a rareness assessment for each vector in the generated training data set. The resulting significance measure can be used to achieve considerably improved classification results for a wide variety of applications by systematically controlling training parameters like learning rate or frequency of presentation for each single vector.
Increasing demands on health care is caused by aging population and passing of legislation such as affordable health care act. It is a well known fact that large number of baby boomers is reaching retirement in the very near future. Many people will be entering nursing homes, retirement homes or prefer to stay home. This along with Affordable health care act creates a huge demand for providing efficient and timely health care for this population. The colleges will not be producing medical doctors, nurses and other professionals fast enough to meet this growing demand. With the increasing availability of sensors, computation and wireless networks, some of these challenges can be met economically and efficiently. The purpose of this paper is to address the issues involved and provide some solutions in meeting these future challenges.
In the reinforcement leaning task, the off-policy algorithms which approximately evaluate the values of states faced with the problem of high evaluation error and were sensitive to the distribution of behavior policy. In order to solve these problems, the basis function optimization method under the off-policy scenario was proposed. The algorithm set the Bellman error of the target policy which was computed with off-policy prediction algorithms as the objective function, then adjust the placement and shape of the basis functions in cooperate with the method of cross-entropy optimization. The experimental results on the grid world show that the algorithm effectively reduced the evaluation error and improved the approximation. Additionally, the algorithm could be easily extended to the problems of large state spaces.
The objective of this paper is to explore the current notions of systems and "System of Systems" and establish the case for quantitative characterization of their structural, behavioural and contextual facets that will pave the way for further formal development (mathematical formulation). This is partly driven by stakeholder needs and perspectives and also in response to the necessity to attribute and communicate the properties of a system more succinctly, meaningfully and efficiently.The systematic quantitative characterization framework proposed will endeavor to extend the notion of emergence that allows the definition of appropriate metrics in the context of a number of systems ontologies. The general characteristic and information content of the ontologies relevant to system and system of system will be specified but not developed at this stage.The current supra-system, system and sub-system hierarchy is also explored for the formalisation of a standard notation in order to depict a relative scale and order and avoid the seemingly arbitrary attributions.
Surrogate modelling based optimization algorithms have been regarded as a powerful tool for solving expensive-to-evaluate functions, and numerous successful applications on optimization problems from various fields have been reported in literature. However, little effort has been devoted to solve complex combinatorial optimization problems through surrogate modelling, since evaluation for solutions of these problems is computationally cheap in general sense. In this paper, we firstly propose a two-layered decomposition of bottleneck stage scheduling problem, in which the subproblem of upper layer can be regarded as an expensive-to-evaluate problem, and the subproblem of lower layer is comparatively easy to solve. Then, we present a differential evolution algorithm combined with a surrogate model to solve the upper-layer subproblem, and the lower-layer subproblem is solved by an effective brand and bound algorithm. Considering that simulation data is generated in a continuous manner, we adopt an incremental extreme learning machine as the surrogate model to reduce the computational cost while preserving good generalization performance. Computational experiments demonstrate the effectiveness and efficiency of the proposed hybrid approach.
Path planning problem is one of the most important and challenging issue in robot control field. In this paper, an improved bioinspired neural network approach is proposed for real-time path planning of robots. In the proposed approach, a new function is used to calculate the connection weight of the bioinspired neural network, to reduce the fluctuation of the path produced by the general bioinspired neural network. Furthermore, a dynamic risk level is introduced into the proposed approach, to improve the performance of the proposed approach in dynamic obstacle avoidance task. In comparison to the general bioinspired neural network based method, experimental results show that the trajectories of robot produced by the proposed approach is optimized, and the proposed approach can deal with the path planning task in dynamic environment efficiently.
A surface EMG signal is one of the most widely used signals as input signals to wearable robots. However, EMG signals that are used to estimate motions are not always available to all users. On the other hand, an EEG signal has drawn attention as input signals for those robots in recent years. The EEG signals can be measured even with amputees and paralyzed patients who are not able to generate some EMG signals. However, the measured EEG signal does not have one-to-one relationships with the corresponding brain part. Therefore, it is more difficult to find the required signals for the control of the robot in accordance with the intention of the user's motion using the EEG signals compared with that using the EMG signals. In this paper, both the EMG and EEG signals are used to estimate the user's motion intention. In the proposed method, the EMG signals are used as main input signals because the EMG signals have higher relative to the motion of a user in comparison with the EEG signals. The EEG signals are used as sub signals in order to cover the estimation of the intention of the user's motion when all required EMG signals cannot be measured. The effectiveness of the proposed method has been evaluated by performing experiments.
We propose an object detection method based on a saliency map using a reference image containing complex background for service robots. In previous detection methods, images that the user prepares in advance contain mostly simple background. However, in order to make robots perform daily tasks, the robots should be able to detect an object using snapshots that contain a complex background. In order to decrease the effect of features in the background, our proposed method classifies local features based on saliency from images. This paper shows the efficacy of the proposed method; furthermore, we demonstrate that our service robot detects certain objects according to the proposed method.
In this paper, the implementation of a discrete-time neural model in an field programmable gate array (FPGA) is proposed to model insulin-glucose dynamics of type 1 diabetes mellitus (T1DM) patients. The neural model is obtained from an on-line neural identifier, which uses a recurrent high-order neural network (RHONN) trained with an extended Kalman filter (EKF), which captures the nonlinear behavior of this dynamics. Experimental data given by continuous glucose monitoring (CGM) device are utilized for identification.
Automobiles are a necessary part of human existence. The utilization of conventional steering for propelling has been a subject of outdated existence from the past century and there is a strong need to shift to better and simpler option. It is widely known that electronic is stated to be the future with imperative of revolutionizing the present day automation by digitalization. In view of the present scenario and subjected needs the work explores and suggests a new technique to simplify driving while ensuring enhanced safety. The proposed system utilizes electronic and touch screen technology and is introduced primarily in cars to corroborate the importance of reducing accidents and making driving enjoyable. The conventional steering is replaced with a touch screen essentially consisting of a panel with multi- utility functions. The use of electronics in steering is a novel concept yet to be explored. A transparent display screen reflects the replica of touch screen. The proposed design utilizes the functional and operational significance of electronics, sensors and cameras in reducing human efforts. The proposed design further ensures the safety with airbags fixed at outskirts of car near the bumper area to make car resistant to accidents. The design retains the conventional steering as its integral part to provide sustainable options for more user friendly driving mode. Practically, the driver in cars with proposed design will have to remain attenuated which will assist in cautious driving. The multi-faceted features of touch screen driving in addition to the one existing will provide safer and enjoyable driving. In a wider domain of automation the proposed design ensures enhanced safety. The requirement of minimum human efforts aided by accurate predictions will make automobile more resistant to accidents. In comparison to the existing automobile systems (cars) being in touch with the current technology on terms of being digitalized every detail provided is precise and strengthens the security levels by multiple folds (makes it theft resistant). The existing design is more feasible and viable to the extent that it can be installed in any car, offers dual mode of driving and requires minimum human efforts. Moreover, the proposed system rests on current technology (of touch) that increases the driving experience and accessibility. As it is linked with electronics and software many existing hi-tech software can be used. Though the proposed design is yet to be implemented, this adaptation of present day touch technology is expected to prove a boon for automobiles and mankind.
