• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.41 no.5
• /
• pp.135-141
• /
• 2004

Image interpolation is an image enlargement method that calculates an empty pixel value using the information of given pixel values. Since a natural image is composed of various spatial frequency components, it is difficult for one method to interpolate pixels with various spatial frequencies. In this paper, we propose an image interpolation method using multiple neural networks with spatial frequency characteristic. Input image is segmented according to spatial frequency by local variance, and each segmented image is interpolated using neural network established for spatial frequency band. The proposed method is applied to line doubling that becomes an important part in image interpolation because of deinterlacing. In simulation the proposed algorithm shows the improved PSNR result compared with conventional algorithms and method using single neural network.

• Smart Media Journal
• /
• v.9 no.4
• /
• pp.17-25
• /
• 2020

In the image object classification problem, low-resolution images may have a negative impact on the classification result, especially when the classification method, such as a convolutional neural network (CNN) model, is trained on a high-resolution (HR) image dataset. In this paper, we analyze the behavior of applying a classical super-resolution (SR) method such as bicubic interpolation, and a deep CNN model such as SRCNN to enhance low-resolution (LR) weeds images used for classification. Using an HR dataset, we first train a CNN model for weeds image classification with a default input size of 128 × 128. Then, given an LR weeds image, we rescale to default input size by applying the bicubic interpolation or the SRCNN model. We analyze these two approaches on the Chonnam National University (CNU) weeds dataset and find that SRCNN is suitable for the image size is smaller than 80 × 80, while bicubic interpolation is convenient for a larger image.

• Journal of Navigation and Port Research
• /
• v.28 no.7
• /
• pp.629-634
• /
• 2004

Continuous depth data are often required in applications of both onboard systems and maritime simulation. But data available are usually discrete and irregularly distributed. Based on the neuron network technique, methods of interpolation to the charted depth are suggested in this paper. Two algorithms based on Levenberg-Marquardt back-propaganda and radial-basis function networks are investigated respectively. A dynamic neuron network system is developed which satisfies both real time and mass processing applications. Using hyperbolic paraboloid and typical chart area, effectiveness of the algorithms is tested and error analysis presented. Special process in practical applications such as partition of lager areas, normalization and selection of depth contour data are also illustrated.

• Proceedings of the KIEE Conference
• /
• 2004.07d
• /
• pp.2528-2530
• /
• 2004

Neural network minimization problems are often conditioned and in this contribution way to handle this will be discussed. It is shown that a better conditioned minimization problem can be obtained if the problem is separated with respect to the linear parameters. This will increase the convergence speed of the minimization. One of the most powerful uses of neural networks is in function approximation(curve fitting)[1]. A main characteristic of this solution is that function (f) to be approximated is given not explicitly but implicitly through a set of input-output pairs, named as training set, that can be easily obtained from calibration data of the measurement system. In this context, the usage of Neural Network(NN) techniques for modeling the systems behavior can provide lower interpolation errors when compared with classical methods like polynomial interpolation. This paper solve of single-variable minimization using neural network.

• Journal of Environmental Impact Assessment
• /
• v.18 no.5
• /
• pp.271-280
• /
• 2009

The objective of this study is to develop a reliable flow forecasting model based on neural network algorithm in order to provide flow rate at stream sections without flow measurement in Nakdong river. Stream flow rate measured at 8-days interval by Nakdong river environment research center, daily upper dam discharge and precipitation data connecting upstream stage gauge were used in this development. Back propagation neural network and multi-layer with hidden layer that exists between input and output layer are used in model learning and constructing, respectively. Model calibration and verification is conducted based on observed data from 3 station in Nakdong river.

• Journal of Applied and Pure Mathematics
• /
• v.4 no.3_4
• /
• pp.185-209
• /
• 2022

Here we exhibit multivariate quantitative approximations of Banach space valued continuous multivariate functions on a box or ℝ^N, N ∈ ℕ, by the multivariate normalized, quasi-interpolation, Kantorovich type and quadrature type neural network operators. We treat also the case of approximation by iterated operators of the last four types. These approximations are achieved by establishing multidimensional Jackson type inequalities involving the multivariate modulus of continuity of the engaged function or its high order Fréchet derivatives. Our multivariate operators are defined by using a multidimensional density function induced by the generalized symmetrical sigmoid function. The approximations are point-wise and uniform. The related feed-forward neural network is with one hidden layer.

• Journal of Applied and Pure Mathematics
• /
• v.5 no.1_2
• /
• pp.69-93
• /
• 2023

Here we give multivariate quantitative approximations of Banach space valued continuous multivariate functions on a box or ℝ^N, N ∈ ℕ, by the multivariate normalized, quasi-interpolation, Kantorovich type and quadrature type neural network operators. We treat also the case of approximation by iterated operators of the last four types. These approximations are derived by establishing multidimensional Jackson type inequalities involving the multivariate modulus of continuity of the engaged function or its high order Fréchet derivatives. Our multivariate operators are defined by using a multidimensional density function induced by a parametrized Gudermannian sigmoid function. The approximations are pointwise and uniform. The related feed-forward neural network is with one hidden layer.

• Korean Journal of Remote Sensing
• /
• v.38 no.2
• /
• pp.167-177
• /
• 2022

In order to obtain satellite image products using the image transmitted to the ground station after capturing the satellite images, many image pre/post-processing steps are involved. During the pre/post-processing, when converting from level 1R images to level 1G images, geometric correction is essential. An interpolation method necessary for geometric correction is inevitably used, and the quality of the level 1G images is determined according to the accuracy of the interpolation method. Also, it is crucial to speed up the interpolation algorithm by the level processor. In this paper, we proposed a lightweight CNN-based interpolation method required for geometric correction when converting from level 1R to level 1G. The proposed method doubles the resolution of satellite images and constructs a deep learning network with a lightweight deep convolutional neural network for fast processing speed. In addition, a feature map fusion method capable of improving the image quality of multispectral (MS) bands using panchromatic (PAN) band information was proposed. The images obtained through the proposed interpolation method improved by about 0.4 dB for the PAN image and about 4.9 dB for the MS image in the quantitative peak signal-to-noise ratio (PSNR) index compared to the existing deep learning-based interpolation methods. In addition, it was confirmed that the time required to acquire an image that is twice the resolution of the 36,500×36,500 input image based on the PAN image size is improved by about 1.6 times compared to the existing deep learning-based interpolation method.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.18 no.7
• /
• pp.907-912
• /
• 1993

In this Paper we have proposed a new method to implement the interpolation of the functions, using a neural network. The architecture of neural network is a three-layer perceptron and the training algorithm is a modified error back propagation algorithm adding neurons to hidden layer. The interpolated functions are sin(7 X), 3rd order polynomial 0.5$\times$3_2$\times$2+X+2.5 and rectangular pulse 0.99 U (X-0.2) -0.99 U(X-0.8) +0.01, where U(X) is the unit step. The root mean squred errors of the interpolated functions are 0.00258, 0.00164 and 0.00116 respectively.

• Proceedings of the KSRS Conference
• /
• 2007.10a
• /
• pp.386-389
• /
• 2007

Currently, fast and accurate long baseline positioning in kinematic mode is a challenging topic, but positional accuracy can be improved with the help of the network-derived external ionospheric corrections. To provide not only ionospheric corrections, but also their variances, satellite-by-satellite interpolation for the ionospheric delays is performed using the least-squares collocation (LSC) method. Satellite-by-satellite interpolation has the advantage in that the vertical projection used in single-layer ionospheric model is not required. Also, more reliable user positioning and the corresponding accuracy assessment can be obtained by providing not only external ionospheric corrections but also their variances. The rover positioning with and without the external ionospheric delays in both rapid-static and kinematic mode was performed and analyzed. The numerical results indicate that the improvement in the positioning quality is achieved using the proposed method. With the TAMDEF network in Antarctica, 18 % improvement in mean time-to-fix in kinematic mode was achieved.