• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.64 no.10
• /
• pp.1469-1478
• /
• 2015

In this paper, The methodology of Type-2 fuzzy set-based Radial Basis Function Neural Network(T2RBFNN) is proposed for Sewage Treatment Process and the simulator is developed for application to the real-world sewage treatment plant by using the proposed model. The proposed model has robust characteristic than conventional RBFNN. architecture of network consist of three layers such as input layer, hidden layer and output layer of RBFNN, and Type-2 fuzzy set is applied to receptive field in contrast with conventional radial basis function. In addition, the connection weights of the proposed model are defined as linear polynomial function, and then are learned through Back-Propagation(BP). Type reduction is carried out by using Karnik and Mendel(KM) algorithm between hidden layer and output layer. Sewage treatment data obtained from real-world sewage treatment plant is employed to evaluate performance of the proposed model, and their results are analyzed as well as compared with those of conventional RBFNN.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.16 no.5
• /
• pp.209-214
• /
• 2021

Panoptic segmentation, which is now widely used in computer vision such as medical image analysis, and autonomous driving, helps understanding an image with holistic view. It identifies each pixel by assigning a unique class ID, and an instance ID. Specifically, it can classify 'thing' from 'stuff', and provide pixel-wise results of semantic prediction and object detection. As a result, it can solve both semantic segmentation and instance segmentation tasks through a unified single model, producing two different contexts for two segmentation tasks. Semantic segmentation task focuses on how to obtain multi-scale features from large receptive field, without losing low-level features. On the other hand, instance segmentation task focuses on how to separate 'thing' from 'stuff' and how to produce the representation of detected objects. With the advances of both segmentation techniques, several panoptic segmentation models have been proposed. Many researchers try to solve discrepancy problems between results of two segmentation branches that can be caused on the boundary of the object. In this survey paper, we will introduce the concept of panoptic segmentation, categorize the existing method into two representative methods and explain how it is operated on two methods: top-down method and bottom-up method. Then, we will analyze the performance of various methods with experimental results.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.26-26
• /
• 2023

General circulation models (GCMs) are widely used in hydrological prediction, however their coarse grids make them unsuitable for regional analysis, therefore a downscaling method is required to utilize them in hydrological assessment. As one of the downscaling methods, convolutional neural network (CNN)-based downscaling has been proposed in recent years. The aim of this study is to generate the process of dynamic downscaling using CNNs by applying GCM output as input and RCM output as label data output. Prediction accuracy is compared between different input datasets, and model structures. Several input datasets with key atmospheric variables such as precipitation, temperature, and humidity were tested with two different formats; one is two-dimensional data and the other one is three-dimensional data. And in the model structure, the hyperparameters were tested to check the effect on model accuracy. The results of the experiments on the input dataset showed that the accuracy was higher for the input dataset without precipitation than with precipitation. The results of the experiments on the model structure showed that substantially increasing the number of convolutions resulted in higher accuracy, however increasing the size of the receptive field did not necessarily lead to higher accuracy. Though further investigation is required for the application, this paper can contribute to the development of efficient downscaling method with CNNs.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.18 no.9
• /
• pp.2642-2664
• /
• 2024

In the process of modern traffic management, information technology has become an important part of intelligent traffic governance. Real-time monitoring can accurately and effectively track and record vehicles, which is of great significance to modern urban traffic management. Existing tracking algorithms are affected by the environment, viewpoint, etc., and often have problems such as false detection, imprecise anchor boxes, and ID switch. Based on the YOLOv5 algorithm, we improve the loss function, propose a new feature extraction module to obtain the receptive field at different scales, and do adaptive fusion with the SGE attention mechanism, so that it can effectively suppress the noise information during feature extraction. The trained model improves the mAP value by 5.7% on the public dataset UA-DETRAC without increasing the amount of calculations. Meanwhile, for vehicle feature recognition, we adaptively adjust the network structure of the DeepSort tracking algorithm. Finally, we tested the tracking algorithm on the public dataset and in a realistic scenario. The results show that the improved algorithm has an increase in the values of MOTA and MT etc., which generally improves the reliability of vehicle tracking.

• Current Optics and Photonics
• /
• v.8 no.5
• /
• pp.463-471
• /
• 2024

Multimode fibers (MMFs) possess high information throughput and small core diameter, making them highly promising for applications such as endoscopy and communication. However, modal dispersion hinders the direct use of MMFs for image transmission. By training neural networks on time-series waveforms collected from MMFs it is possible to reconstruct images, transforming blurred speckle patterns into recognizable images. This paper proposes a fully convolutional neural-network model, MSMDFNet, for image restoration in MMFs. The network employs an encoder-decoder architecture, integrating multiscale convolutional modules in the decoding layers to enhance the receptive field for feature extraction. Additionally, attention mechanisms are incorporated from both spatial and channel dimensions, to improve the network's feature-perception capabilities. The algorithm demonstrates excellent performance on MNIST and Fashion-MNIST datasets collected through MMFs, showing significant improvements in various metrics such as SSIM.

• Restorative Dentistry and Endodontics
• /
• v.27 no.6
• /
• pp.587-599
• /
• 2002

Extracellular single unit recordings were made from the ventral posteromedial thalamic (VPM) nociceptive neurons to determine mechanoreceptive field (RF) and response properties. A total of 44 VPM thalamic nociceptive neurons were isolated from rats anesthetized with urethane-chloralose. Based on responses to various mechanical stimuli including touch, pressure and pinch applied to the RF, 32 of 44 neurons were classified as nociceptive specific (NS) neuron. The other 12 neurons, classified as wide dynamic range (WDR), showed a graded response to increasingly intense stimuli, with a maximum discharge to noxious pinch. The VPM nociceptive neurons showed various spontaneous activity ranged from 0-6 Hz. They were located throughout the VPM, and had an contralateral RF including mainly intraoral (and perioral) regions. The RF size was relatively small, and very few neurons had a receptive field involving 3 trigeminal divisions. The NS neurons activated only by pressure and pinch stimuli had high mechanical thresholds compared to WDR neurons activated also by touch stimuli. The VPM nociceptive neurons were tested with suprathershold graded mechanical stimuli. Most of 21 NS and 8 WDR neurons showed a progressive increase in number of spikes as mechanical stimulus intensity was increased. In some neurons, the responses reached a peak before the highest intensity was given. Application of 5 mM $CoCl_2{\;}(10{\;}{\mu}\ell)$ solution to the trigeminal subnucleus caudalis did not produce any significant changes in the spontaneous activity, RF size, mechanical threshold, and response to suprathreshold mechanical stimuli of 9 VPM nociceptive neurons tested. 17 of 33 VPM nociceptive neurons responded to noxious heat as well as noxious mechanical stimuli applied to their RF. Application of the mustard oil, a small-fiber excitant and inflammatory irritant, to the right maxillary first molar tooth pulp induced an immediate but short-lasting neuronal discharges upto approximately 4 min in 16 of 42 VPM nociceptive neurons. These results suggest that VPM thalamic nucleus may contribute to the sensory discriminative aspect of orofacial nociception.

• Restorative Dentistry and Endodontics
• /
• v.24 no.4
• /
• pp.614-622
• /
• 1999

Recent studies have implicated that more rostral components of the trigeminal spinal nucleus including subnucleus oralis (Vo) in orofacial nociceptive mechanisms. Since there is only limited electrophysiological evidence, the present study was initiated to characterize the receptive field and response properties of malls nociceptive neurons in chloralose/urethan-anesthetized rats. Single neuronal activity was recorded in right subnucleus oralis, and types of nociceptive neurons classified wide dynamic range (WDR), NS (nociceptive specific) and deep nociceptive (D) and the mechanoreceptive field (RF) and response properties were determined. A total of 34 nociceptive neurons could be subdivided into 17WDR neurons, 12NS neurons and 5D neurons. Vo nociceptive neurons had RF involving maxillary and/or mandibular divisions mainly located in the intraoral and/or perioral regions. Majority of Vo nociceptive neurons showed spontaneous activity less than 1Hz. The NS and D neurons activated only by heavy pressure and/or pinch stimuli had high mechanical thresholds compared to WDR neurons activated also by tactile stimuli. Vo nociceptive neurons showed a progressive increase of response to the graded mechanical stimuli. 39% of Vo nociceptive neurons received C-fiber electrical input as well as A-fiber electrical input from their RF, and 45% of them responded to electrical stimulation of the right maxillary first molar. 41% of Vo nociceptive neurons responded to noxious heat applied to their RF, and 18% of them showed an immediate burst of discharges following MO application to the right maxillary first molar pulp. These results indicate that Vo is involved in the transmission of nociceptive information mainly coming from intraoral or perioral region including tooth pulp.

• Progress in Medical Physics
• /
• v.9 no.2
• /
• pp.95-104
• /
• 1998

The dynamic properties of the 3rd-order neuron of the retina was investigated by using conventional intracellular recording techniques. Experiments were performed in the superfused retina-eyecup preparation of the channel catfish, Ictalurus punctatus. The cornea, iris, lens, and vitreous were removed by absorption with Kimwipe tissue under the dissection microscope thereby exposing the retina in a hemi -eyecup. The electrical signal was amplified by electrometer, viewed on oscilloscope. Regular signals from the cells were recorded on a penwriter and stored by data recorder and computer. Full-field, spot or annular light stimuli were generated on a computer monitor and focused onto the retina. Baclofen hyperpolarized the dark membrane potential, suppressed sustained component and enhanced transient component of the ON-sustained cell with a large transient component, but did not affect the surround antagonism of the cell. Baclofen selectively suppressed responses evoked by moving bar light stimuli on the ON-OFF transient cell. The results suggest that transient cells have directional selectivity in the inner retina. These dynamic properties of amacrine and ganglion cells were modulated by baclofen. Therefore, it is presumed that there is baclofen-induced directional selectivity in ON-OFF transient cells in the catfish retina.

• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Journal of Korean Neurosurgical Society
• /
• v.29 no.8
• /
• pp.985-994
• /
• 2000

Objective : There are some advantages of trigeminal evoked potential(TEP) recording compared to other somatosensory evoked potential(SSEP) recordings. The trigeminal sensory pathway has a pure sensory nerve branch, a broader receptive field in cerebral cortex, and a shorter pathway. Despite these advantages, there is little agreement as to what constitutes a normal response and what wave forms truly characterize the intraoperative TEP. This study presents the normative data of TEP recorded on the epidural surface of the rat with a platinum ball electrode. Materials & Methods : Under general anesthesia with urethane, the adult Sprague-Dawley male rats(300-350g) were given electrical stimulation with two stainless steel electrodes which were inserted into the subcutaneous layer of the area around whiskers. A reference electrode was positioned in the temporalis muscle ipsilateral to the recording site. Results : TEPs were recorded in the Par I area of somatosensory cortex and recorded most apparently on the point of 2mm posterior from the bregma and 6mm lateral from the midline. The typical wave form consisted of 5 peaks (N1-P1-N2-P2-N3 according to emerging order, upward negativity). Each latency to corresponding peaks was not influenced by the different intensities of stimulation, especially from 1 to 5mA. Average latencies of 5 peaks were in the following order ; 7.7, 11.1, 15, 22.3, 29.4ms. There was also no significant difference between latencies before and after administration of muscle relaxant(pancuronium). For the electrophysiological localization of recorded waves, the action potential of a single unit was recorded with glass microelectrode(filled with 2M NaCl, $3-5M{\Omega}$) in the thalamus of rat. A sharp wave was recorded in the VPM nucleus, in which the latency was shorter than that of N1. This suggests that all 5 peaks were generated by neural activities in the suprathalamic pathway. Conclusion : In terms of recording near-field potentials, our data also suggests that TEP in the rat may be superior to other SSEPs. In overall, these results may afford normative data for the studies of supratentorial lesions such as hydrocephalus or cerebral ischemia which can have an influence on near-field potentials.