• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.11 no.11
• /
• pp.2143-2149
• /
• 2007

Visual Cortex Stimulator is among artificial retina prosthesis for blind man, is the method that stimulate the brain cell directly without processing the information from retina to visual cortex. In this paper, we propose image construction and recognition model that is similar to human visual processing by recognizing the feature data with orientation information, that is, the characteristics of visual cortex. Back propagation algorithm based on Delta-bar delta is used to recognize after extracting image feature by Kirsh edge detector. Various numerical patterns are used to analyze the performance of proposed method. In experiment, the proposed recognition model to extract image characteristics with the orientation of information from retinal cells to visual cortex makes a little difference in a recognition rate but shows that it is not sensitive in a variety of learning rates similar to human vision system.

• Journal of Digital Convergence
• /
• v.12 no.11
• /
• pp.349-355
• /
• 2014

In this paper, we propose the artificial vision system using human visual information processing and wavelet. Artificial vision system may be used for the visually impaired person and the machine recognition system. In this paper, we have constructed the information compression process to ganglion cells from the human retina. And we have reconstructed the primary visual information using recovery process to primary visual cortex from ganglion. Primary visual information is constructed by wavelet transformation using a high frequency and low frequency response. In the experiment, we used the faces database of AT&T. And the proposed method was able to improve the accuracy of face recognition considerably. And it was verified through experiments.

• Journal of Korea Multimedia Society
• /
• v.22 no.3
• /
• pp.349-356
• /
• 2019

When a prescription change occurs in the hospital depending on a patient's improvement status, pharmacists directly classify manually returned pills which are not taken by a patient. There are hundreds of kinds of pills to classify. Because it is manual, mistakes can occur and which can lead to medical accidents. In this study, we have compared YOLO, Faster R-CNN and RetinaNet to classify and detect pills. The data consisted of 10 classes and used 100 images per class. To evaluate the performance of each model, we used cross-validation. As a result, the YOLO Model had sensitivity of 91.05%, FPs/image of 0.0507. The Faster R-CNN's sensitivity was 99.6% and FPs/image was 0.0089. The RetinaNet showed sensitivity of 98.31% and FPs/image of 0.0119. Faster RCNN showed the best performance among these three models tested. Thus, the most appropriate model for classifying pills among the three models is the Faster R-CNN with the most accurate detection and classification results and a low FP/image.

• Journal of Biomedical Engineering Research
• /
• v.28 no.2
• /
• pp.212-217
• /
• 2007

As a preliminary study for the development of electrical stimulation strategy of artificial retina, we set up a method fur the reconstruction of input intensity variation from retinal ganglion cell(RGC) responses. In order to estimate light intensity variation, we used an optimal linear filter trained from given stimulus intensity variation and multiple single unit spike trains from RGCs. By applying ON/OFF stimulation(ON duration: 2 sec, OFF duration: 5 sec) repetitively, we identified three functional types of ganglion cells according to when they respond to the ON/OFF stimulus actively: ON cell, OFF cell, and ON-OFF cell. Experiments were also performed using a Gaussian random stimulus and a binary random stimulus. The input intensity was updated once every 90 msec(i. e. 11 Hz) to present the stimulus. The result of reconstructing 11 Hz Gaussian and binary random stimulus was not satisfactory and showed low correlation between the original and reconstructed stimulus. In the case of ON/OFF stimulus in which temporal variation is slow, successful reconstruction was achieved and the correlation coefficient was as high as 0.8.

• Journal of Life Science
• /
• v.17 no.5 s.85
• /
• pp.634-640
• /
• 2007

Our previous study suggested that n-3 fatty acid deficiency was associated with significantly reduced spatial learning as assessed by Morris water maze test. Here we investigated an effect of n-3 fatty acid deficiency on rat brain, retina and serum fatty acyl compositions at 15 wks age using a first generational artificial rearing technique. Newborn Rat pups were separated on day 2 and assigned to two artificial rearing groups or a dam-reared control group. Pups were hand fed artificial milk via custom-designed nursing bottles containing either 0.02%(n-3 Deficient) or 3.1% (n-3 Adequate) of total fatty acids as a-linolenic acid(LNA). At day 21, rats were weaned to either n-3 deficient or n-3 adequate pelleted diets and fatty acid compositions of brain, retina and liver were analyzed at 15 wks age. Brain docosahexaenoic acid(DHA) was lower(58% and 61%, P<0.05) in n-3 deficient in comparison to n-3 adequate and dam-reared groups, receptively, while brain docosapentaenoic acid(DPAn-6) was increased in the n-3 deficient group. In retina and serum fatty acid compositions, the decreased precentage of DHA and increased precentage of DPAn-6 were observed. These results suggested that artificial rearing method can be used to produce n-3 fatty acid deficiency in the first generation and that adequate brain DHA levels are required for optimal brain function.

• Annual Conference of KIPS
• /
• 2020.05a
• /
• pp.446-450
• /
• 2020

딥러닝 기반의 영상 분석 방법들은 많은 양의 학습 데이터가 필요하며, 학습 데이터 구축에는 많은 시간과 노력이 소요된다. 특히 객체 검출 분야의 경우 영상 내 객체의 위치, 크기, 범주 등의 정보가 모두 필요하여 학습 데이터 구축에 더 많은 어려움이 있으며, 이를 해결하기 위해 최근 이미지 합성기반 데이터 증강에 관한 연구가 활발히 진행되고 있다. 이미지 합성기반 데이터 증강 방법은 배경 영상에 객체를 합성할 때 객체와 배경 영상이 접한 영역에서 아티팩트(Artifact)가 발생하며, 이는 객체 검출 모델이 아티팩트를 객체의 특징으로 모델링하여 검출 성능이 저하되는 원인이 된다. 이러한 문제를 해결하기 위하여 본 논문에서는 양방향 필터 기반의 이미지 합성 방법을 제안하고, 단일 단계 검출의 대표적인 방법인 RetinaNet을 이용하여 이미지 합성기반 데이터 증강 방법의 성능을 분석하였다. 공개 데이터셋에 대한 실험 결과 본 논문에서 사용한 단일 검출 방법 및 데이터 증강 기법을 사용하면 더 적은 양의 증강 데이터로 기존 방법과 동일한 성능을 보여주는 것을 확인하였다.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.40 no.5
• /
• pp.381-391
• /
• 2022

This paper is intended to find one of the prevailing deep learning models that are a type of AI (Artificial Intelligence) that helps rapidly detect damaged buildings where disasters occur. The models selected are SSD-512, RetinaNet, and YOLOv3 which are widely used in object detection in recent years. These models are based on one-stage detector networks that are suitable for rapid object detection. These are often used for object detection due to their advantages in structure and high speed but not for damaged building detection in disaster management. In this study, we first trained each of the algorithms on xBD dataset that provides the post-disaster imagery with damage classification labels. Next, the three models are quantitatively evaluated with the mAP(mean Average Precision) and the FPS (Frames Per Second). The mAP of YOLOv3 is recorded at 34.39%, and the FPS reached 46. The mAP of RetinaNet recorded 36.06%, which is 1.67% higher than YOLOv3, but the FPS is one-third of YOLOv3. SSD-512 received significantly lower values than the results of YOLOv3 on two quantitative indicators. In a disaster situation, a rapid and precise investigation of damaged buildings is essential for effective disaster response. Accordingly, it is expected that the results obtained through this study can be effectively used for the rapid response in disaster management.

• The Korean Journal of Physiology and Pharmacology
• /
• v.27 no.6
• /
• pp.541-553
• /
• 2023

Retinal prostheses have shown some clinical success in restoring vision in patients with retinitis pigmentosa. However, the post-implantation visual acuity does not exceed that of legal blindness. The reason for the poor visual acuity might be that (1) degenerate retinal ganglion cells (RGCs) are less responsive to electrical stimulation than normal RGCs, and (2) electrically-evoked RGC spikes show a more widespread not focal response. The single-biphasic pulse electrical stimulation, commonly used in artificial vision, has limitations in addressing these issues. In this study, we propose the benefit of multiple consecutive-biphasic pulse stimulation. We used C57BL/6J mice and C3H/HeJ (rd1) mice for the normal retina and retinal degeneration model. An 8 × 8 multi-electrode array was used to record electrically-evoked RGC spikes. We compared RGC responses when increasing the amplitude of a single biphasic pulse versus increasing the number of consecutive biphasic pulses at the same stimulus charge. Increasing the amplitude of a single biphasic pulse induced more RGC spike firing while the spatial resolution of RGC populations decreased. For multiple consecutive-biphasic pulse stimulation, RGC firing increased as the number of pulses increased, and the spatial resolution of RGC populations was well preserved even up to 5 pulses. Multiple consecutive-biphasic pulse stimulation using two or three pulses in degenerate retinas induced as much RGC spike firing as in normal retinas. These findings suggest that the newly proposed multiple consecutive-biphasic pulse stimulation can improve the visual acuity in prosthesis-implanted patients.

• Proceedings of the Korean Information Science Society Conference
• /
• 2007.06c
• /
• pp.436-439
• /
• 2007

Noise reduction in the image is very important to improve the quality of the image. This paper discusses a new noise reduction technique which uses the On/Off spatio-temporal structure of the receptive fields. Also this paper proposes a structurally improved artificial vision system which incorporates the sphere type retina structure, an improved On/Off spatio-temporal receptive fields structure, and chiasm for hemianopia testing.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06d
• /
• pp.369-371
• /
• 2012

Wireless sensor networks have taken immense interest in healthcare systems in recent years. One example of it is in an in vivo sensor that is deployed in critical and sensitive healthcare applications like artificial retina, cardiac pacemaker, drug delivery, blood pressure, internal heat calculation, glucosemonitoring etc. In vivo sensor nodes exhibit temperature that may be very dangerous for human tissues. However, existing in vivo thermal aware routing approaches suffer from hotspot creation, delay, and computational complexity. These limitations motivate us toward an in vivo virtual backbone, a small subset of nodes, connected to all other nodes and involved in routing of all nodes, -based solution. A virtual backbone is lightweight and its fault-tolerant version allows in vivo sensor nodes to disconnect hotspot paths and to use alternative paths. We have formulated the problem as m-connected k-dominating set problem with minimum temperature cost in in vivo sensor network. This is a combinatorial optimization problem and we have been motivated to use evolutionary approach to solve the problem.