Search

Most studies on convolutional Neural Network (CNN) based image processing have proposed networks that can be optimized for a single level. Here, the term "level" refers to the specific objective defined for each task, such as the degree of noise in denoising tasks. Hence, they underperform on other levels and must be retrained to deliver optimal performance. Using multiple models to cover multiple levels involves very high computational costs. To solve these problems, recent approaches train the networks on two different levels and propose their own modulation methods to enable the arbitrary intermediate levels. However, many of them 1) have difficulty adapting from one level to the other, 2) suffer from unintended artifacts in the intermediate levels, or 3) require large memory and computational cost. In this paper, we propose a novel framework using Filter Transition Network (FTN), which is a non-linear module that easily adapts to new levels, is regularized to prevent undesirable side-effects, and extremely lightweight being a data-independent module. Additionally, for stable learning of FTN, we newly propose a method to initialize nonlinear CNNs with identity mappings. Extensive results for various image processing tasks indicate that the performance of FTN is stable regarding adaptation and modulation and is comparable to that of the other heavy frameworks. [ABSTRACT FROM AUTHOR]

In Korea, weather forecasts for fundamental weather factors, such as temperature, precipitation, wind direction and speed, humidity, and cloudiness, are provided for a three-day period in each region. This can facilitate predicting photovoltaic power generation based on weather forecasting. For this purpose, in the present paper, we aim to propose corresponding model. However, the Korea Meteorological Administration does not forecast the amount of solar radiation and sunshine that mostly influence the results of photovoltaic power generation prediction. In this study, we predict these parameters considering various input/output (I/O) variables and learning algorithms applied to weather forecasts on hourly weather data. Finally, we predict photovoltaic power generation based on the best sunshine and solar radiation prediction results. The data structure underlying all predictions relies on four models applied to fundamental weather factors on sunshine and solar radiation data two hours ago. Then, the photovoltaic power generation prediction is implemented using four models depending on whether to add the predicted sunshine and solar radiation data obtained at the previous step. The prediction algorithm relies on an adaptive neuro-fuzzy inference system and artificial neural network (ANN) techniques, including dynamic neural network (DNN), recurrent neural network (RNN), and long short-term memory (LSTM). The results of the conducted experiment indicate that ANN perform better than the neuro-fuzzy approach. Moreover, we demonstrate that RNN and LSTM are more suitable for the time series data structures compared with DNN. Furthermore, we report that the weather forecast structure and the model 4 structure, which includes sunshine and solar radiation data two hours ago, achieve the best prediction results. [ABSTRACT FROM AUTHOR]