• Journal of Advanced Navigation Technology
• /
• v.28 no.2
• /
• pp.238-245
• /
• 2024

In this paper, we propose a framework for improving the performance of semantic segmentation of agricultural multispectral image using feature fusion techniques. Most of the semantic segmentation models being studied in the field of smart farms are trained on RGB images and focus on increasing the depth and complexity of the model to improve performance. In this study, we go beyond the conventional approach and optimize and design a model with multispectral and attention mechanisms. The proposed method fuses features from multiple channels collected from a UAV along with a single RGB image to increase feature extraction performance and recognize complementary features to increase the learning effect. We study the model structure to focus on feature fusion and compare its performance with other models by experimenting with favorable channels and combinations for crop images. The experimental results show that the model combining RGB and NDVI performs better than combinations with other channels.

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.502-507
• /
• 2003

Toward the development of practical methods for observed data oriented bispectral estimation, an automatic means for determining the smoothing bandwidth of bispectral windows is proposed, that can also provide an associated optimum bispectral estimate of stationary non-Gaussian signals, systematically only from an observed time series datum of finite length. For the conventional non-parametric bispectral estimation, the MSE (mean squared error) of the normalized estimate is reviewed under a certain mixing condition and sufficient data length, mainly from the viewpoint of the inverse relation between its bias and variance with respect to the smoothing bandwidth. Based on the fundamental relation, a systematic method not only for determining the bandwidth, but also for obtaining the optimum bispectral estimate is presented by newly introducing a MSE evaluation index of the estimate only from an observed time series datum of finite length. The effectiveness and fundamental features of the proposed method are illustrated by the basic results of numerical experiments.

• The Journal of the Acoustical Society of Korea
• /
• v.33 no.4
• /
• pp.268-272
• /
• 2014

This paper presents a method for speech recognition using multi-section vector-quantization (MSVQ) and time-delay recurrent neural network (TDTNN). The MSVQ generates the codebook with normalized uniform sections of voice signal, and the TDRNN performs the speech recognition using the MSVQ codebook. The TDRNN is a time-delay recurrent neural network classifier with two different representations of dynamic context: the time-delayed input nodes represent local dynamic context, while the recursive nodes are able to represent long-term dynamic context of voice signal. The cepstral PLP coefficients were used as speech features. In the speech recognition experiments, the MSVQ/TDRNN speech recognizer shows 97.9 % word recognition rate for speaker independent recognition.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.12 no.6
• /
• pp.108-115
• /
• 2013

A developed skill of an intelligent CCTV is also advancing by using its Image Acquisition Device. In this field, area for technique can be divided into Foreground Subtraction which detects individuals and objects in a potential observing area and a tracing technology which figures out moving route of individuals and objects. In this thesis, an improved algorism for a settled engine development, which is stable to change in both noise and illumination for detecting moving objects is suggested. The proposed algorism from this thesis is focused on designing a stable and real time processing method which is perfect model in detecting individuals, animals, and also low-speeding transports and catching a change in an illumination and noise.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.35 no.1
• /
• pp.23-30
• /
• 2017

NDVI (Normalized Difference Vegetation Index) plays an important role in surface land cover classification and LST (Land Surface Temperature Extraction). Its characteristics do not full carry the information of the surface cover typically in urban areas even though it is widely used in analyses in urban areas as well as in vegetation. However, abnormal NDVI values are frequently found in urban areas. We, therefore, examined NDVI values on whether NDVI is appropriate for LST and whether there are considerations in NDVI analysis typically in urban areas because NDVI is strongly related to the surface emissivity calculation. For the study, we observed the influence of the surface settings (i.e., geometric shape and color) on NDVI values in urban area and transition features between three land cover types, vegetation, urban materials, and water. Interestingly, there were many abnormal NDVI values systematically derived by the surface settings and they might influence on NDVI and eventually LST. Also, there were distinguishable transitions based on the mixture of three surface materials. A transition scenario was described that there are three transition types of mixture (urban material-vegetation, urban material-water, and vegetation-water) based on the relationship of NDVI and LST even though they are widely distributed.

• Journal of The Korean Society of Clinical Toxicology
• /
• v.14 no.2
• /
• pp.71-77
• /
• 2016

Purpose: No studies have been conducted to investigate the acute toxicity of aryloxyphenoxypropionate herbicides in humans following ingestion. Therefore, this study was conducted to investigate the clinical characteristics of aryloxyphenoxypropionate herbicide poisoning and provide guidance for physicians treating patients who have ingested these types of herbicides. Methods: A retrospective observational case series was conducted using ten patients with history of aryloxyphenoxy propionate herbicide. Data were collected for clinical manifestation, management and final outcome. Results: The most common symptoms were gastrointestinal irritation and an altered mental state (Glasgow Coma Scale<15). An elevated lactate level was a common laboratory abnormality, and prolonged QTc interval was commonly observed. These clinical features normalized within one day of supportive treatment. Conclusion: The acute toxicity of aryloxyphenoxypropionate herbicides in humans is manageable with supportive treatment. However, physicians should take into account depressed consciousness, the possibility of arrhythmia, and an elevated lactate level when planning their treatment strategy.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.28 no.6
• /
• pp.816-825
• /
• 2004

In this work, effects of hyper-elastic rubber material properties on the indentation load-deflection curve and subindenter deformation are examined via finite element (FE) analyses. An optimal location for data analysis is selected, which features maximum strain energy density and negligible frictional effect. We then contrive two normalized functions, which map an indentation load vs. deflection curve into a strain energy density vs. first invariant curve. From the strain energy density vs. first invariant curve, we can extract the rubber material properties. This new spherical indentation approach produces the rubber material properties in a manner more effective than the common uniaxial tensile/com-pression tests. The indentation approach successfully measures the rubber material properties and the corresponding nominal stress-strain curve with an average error less than 3%.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.6
• /
• pp.2612-2633
• /
• 2020

Due to the explosive growth of multimedia speech data, how to protect the privacy of speech data and how to efficiently retrieve speech data have become a hot spot for researchers in recent years. In this paper, we proposed an encrypted speech retrieval scheme based on long short-term memory (LSTM) neural network and deep hashing. This scheme not only achieves efficient retrieval of massive speech in cloud environment, but also effectively avoids the risk of sensitive information leakage. Firstly, a novel speech encryption algorithm based on 4D quadratic autonomous hyperchaotic system is proposed to realize the privacy and security of speech data in the cloud. Secondly, the integrated LSTM network model and deep hashing algorithm are used to extract high-level features of speech data. It is used to solve the high dimensional and temporality problems of speech data, and increase the retrieval efficiency and retrieval accuracy of the proposed scheme. Finally, the normalized Hamming distance algorithm is used to achieve matching. Compared with the existing algorithms, the proposed scheme has good discrimination and robustness and it has high recall, precision and retrieval efficiency under various content preserving operations. Meanwhile, the proposed speech encryption algorithm has high key space and can effectively resist exhaustive attacks.

• Proceedings of the IEEK Conference
• /
• 1999.06a
• /
• pp.583-586
• /
• 1999

In this paper, the wavelet transform is performed in the input 256$\times$256 color image and decomposes a image into low-pass and high-pass components. Since the high-pass band contains the components of three directions, edges are detected by combining three parts. After finding the position of face using the histogram of the edge component, a face region in low-pass band is cut off. Since RGB color image is sensitively affected by luminances, the image of low pass component is normalized, and a facial region is detected using face color informations. As the wavelet transform decomposes the detected face region into three layer, the dimension of input image is reduced. In this paper, we use the 3000 images of 10 persons, and KL transform is applied in order to classify face vectors effectively. FCM(Fuzzy C-Means) algorithm classifies face vectors with similar features into the same cluster. In this case, the number of cluster is equal to that of person, and the mean vector of each cluster is used as a codebook. We verify the system performance of the proposed algorithm by the experiments. The recognition rates of learning images and testing image is computed using correlation coefficient and Euclidean distance.

• Proceedings of the KSRS Conference
• /
• 2008.10a
• /
• pp.224-227
• /
• 2008

Land cover and its changes, affecting multiple aspects of the environmental system such as energy balance, biogeochemical cycles, hydrological cycles and the climate system, are regarded as critical elements in global change studies. Especially in arid and semiarid regions, the observation of ecosystem that is sensitive to climate change can improve an understanding of the relationships between climate and ecosystem dynamics. The purpose of this research is analyzing the ecosystem surrounding the Gobi desert in North Asia quantitatively as well as qualitatively more concretely. We used Normalized Difference Vegetation Index (NDVI) derived from SPOT-VEGETATION (VGT) sensor during 1999${\sim}$2007. Ecosystem monitoring of this area is necessary because it is a hot spot in global environment change. This study will allow predicting areas, which are prone to the rapid environmental change. Eight classes were classified and compare with MODerate resolution Imaging Spectrometer (MODIS) global land cover. The time-series analysis was carried out for these 8 classes. Class-1 and -2 have least amplitude variation with low NDVI as barren areas, while other vegetated classes increase in May and decrease in October (maximum value occurs in July and August). Although the several classes have the similar features of NDVI time-series, we detected a slight difference of inter-annual variation among these classes.