Tea Classification Based on Artificial Olfaction Using Bionic Olfactory Neural Network

Yang, Xinling; Fu, Jun; Lou, Zhengguo; Wang, Li-Yu; Li, Guang; Freeman, Walter

doi:10.1007/11760023_50

Cited by 18 publications

(10 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…[4] is that these neural signals exist by virtue of the interaction of population of neurons, forming macrostates in neural ensembles at lower levels. The K3 is constructed from a number of other networks, with at least three layers formed by K2-sets, and has been used successfully in face detection [5] [8], robot navigation [9], recognition of spoken digits [10], [11], classification of structures in tissues [12], teas [13] and text [14], prediction of time series [15], motor imagery [16], and clustering [17]. It has also been used successfully for EEG classification [18]- [22].…”

Section: K3-setmentioning

confidence: 99%

A simulator for Freeman K-sets in Java

Piazentin

Rosa

2015

2015 International Joint Conference on Neural Networks (IJCNN)

View full text Add to dashboard Cite

Abstract-Freeman K-sets form a hierarchy of models of the dynamics of neuron populations at the mesoscopic (intermediate) level. The topology of connections is modeled by networks of excitatory and inhibitory populations of neurons; the dynamics is approximated by piecewise Iinearization of nonlinear ordinary differential equations (ODE). A simulator for the K-sets of levels Ko to K3 was built in Java in order to remedy some of the deficiencies found in the simulator available built in MATLAB, as defects that cause instabilities in the simulation, the lack of flexibility in the parameterization of the K3-set and the high computational time required to use the simulator due to design and language decisions. Java was chosen because it is a popular, multi-platform, object oriented and computationally efficient language. The software developed in Java also allows easy integration with MATLAB and other environments. An application on clustering of social networking data evaluates the proposed simulator.

show abstract

Section: K3-setmentioning

confidence: 99%

A simulator for Freeman K-sets in Java

Piazentin

Rosa

2015

2015 International Joint Conference on Neural Networks (IJCNN)

View full text Add to dashboard Cite

show abstract

“…01, B). More importantly when the KIII set is expanded into multiple 2-D layers of the several interactive KIe and KIi layers, the model performs pattern classification with exceptionally high levels of success, beginning with olfactory stimuli such as kinds of tea, Yang et al, 2006), and extending to quality control of machine screws in industry (Yao et al, 1991), Japanese vowel sounds , EEG patterns (Kozma and Freeman, 2002), face recognition (Li et al, 2004), all 36 alphanumeric characters with an 8x8 array , and EEG patterns in seizure (Ruiz et al, 2009) and hypoxia (Hu et al, 2006), outperforming conventional neural networks in requiring small training sets (typically 5 to 10 examples), one-step convergence by phase transition rather than asymptotic approach by gradient descent, and memory capacity increasing geometrically with the number of nodes (Kozma and Freeman, 2001;Kozma et al, 2005;Ilin, Kozma and Werbos, 2008). Further uses of multisensory pattern classification are achieved by conjoining three KIII sets to form a KIV set, which simulates the dynamics of goaldirected behavior of primitive vertebrates by constructing multisensory percepts (gestalts) for making decisions (Kozma, Freeman and Erdí, 2003) and guiding navigation .…”

Section: A Applications Of Noise-modulated Deterministic Chaosmentioning

confidence: 99%

Chaotic Neocortical Dynamics

Freeman

2013

Chaos, CNN, Memristors and Beyond

Self Cite

View full text Add to dashboard Cite

The first step of the sensory systems is to construct the meaning of the information they receive from the senses. They do this by generating random noise and then filtering the noise with adaptive filters. We simulate the operation with the solutions of matrices of ordinary differential equations that predict bifurcations between point and limit cycle attractors. The second step is integration of the outputs from the several sensory systems into a multisensory percept (gestalt), which in the third step is consolidated and stored as knowledge. Simulation of the second step requires use of landscapes of non-convergent chaotic attractors. This is not deterministic chaos, which is much too brittle owing to the infinite sensitivity to initial conditions. It is a hybrid form we call stochastic chaos, which is stabilized by additive noise modeled on noise sources in the sensory systems. Thus bifurcation and chaos theory provides tools for succinct empirical models of cortical dynamics performing the most basic cognitive operations: generalization, abstraction, and categorization in constructing knowledge. The descriptions are in a form that is suitable for more advanced modeling using analog VLSI, neuropercolation from random graph theory, nonequilibrium dissipative thermodynamics, and macroscopic many-body physics. This review concludes with a summary of the applications of stochastic chaos in pattern classification and some prescriptions for neurobiologists on what to look for in large-scale anatomical formations.

show abstract

“…Yang et al [45] studied the properties of the model in time, frequency and in the time-frequency domains. Recently, the KIII model has been applied to pattern recognition, including odor recognition in eNose [46][47][48][49] . In this section, we review the studies of several biologically oriented learning rules and pattern extraction methods for the KIII model, and their applications to eNose for concentration influence elimination, sensor drift counteraction, odor contrast enhancement and background suppression.…”

Section: Bionic Engineeringmentioning

confidence: 99%

Progress in bionic information processing techniques for an electronic nose based on olfactory models

Zheng

2009

Chin. Sci. Bull.

Self Cite

View full text Add to dashboard Cite

As a novel bionic analytical technique, an electronic nose, inspired by the mechanism of the biological olfactory system and integrated with modern sensing technology, electronic technology and pattern recognition technology, has been widely used in many areas. Moreover, recent basic research findings in biological olfaction combined with computational neuroscience promote its development both in methodology and application. In this review, the basic information processing principle of biological olfaction and artificial olfaction are summarized and compared, and four olfactory models and their applications to electronic noses are presented. Finally, a chaotic olfactory neural network is detailed and the utilization of several biologically oriented learning rules and its spatiotemporal dynamic propties for electronic noses are discussed. The integration of various phenomena and their mechanisms for biological olfaction into an electronic nose context for information processing will not only make them more bionic, but also perform better than conventional methods. However, many problems still remain, which should be solved by further cooperation between theorists and engineers.artificial olfaction, olfactory model, pattern recognition, electronic nose, bionics Various sensors derive different types of information from complicated environments, following which computers analyze before actuators perform particular tasks. To a certain degree, such devices have extended human sensory, brain and physical power to enhance the capacity of understanding and transforming nature. However, biological systems derived from evolution are useful models for engineering design, particularly with regard to bionics.Electronic noses (eNose) are a kind of novel bionic analytical instrument for mimicking the principle of biological olfaction. Generally, an electronic nose consists of an array of cross-sensitive gas sensors and an appropriate pattern recognition method, capable of automatically detecting and discriminating simple or complex odors [1] . Only twenty years after Persaud et al.'s pioneer work [2] , eNoses have been widely utilized within the food industry, for environmental monitoring, public security and medical diagnosis [3,4] , as they tend to be relatively fast, easy to use and objective with a modest overall cost.As a multidisciplinary research field, most studies of eNose primarily focus on gas sensor technology [5] and pattern recognition methods [6] . Pattern recognition models remain inadequately investigated such that they are the bottle-neck for commercialization. With more novel sensing techniques are applied to eNoses [3,7] , signal processing and pattern recognition methods have increased in significance. The information processing mechanism in olfaction from the genetic, cellular to systemic levels is more fully understand as a result of cross-

show abstract

Tea Classification Based on Artificial Olfaction Using Bionic Olfactory Neural Network

Cited by 18 publications

References 9 publications

A simulator for Freeman K-sets in Java

A simulator for Freeman K-sets in Java

Chaotic Neocortical Dynamics

Progress in bionic information processing techniques for an electronic nose based on olfactory models

Contact Info

Product

Resources

About