Accepted Papers

  • A Wind Power Prediction Method based on Bayesian Fusion
    Jianqi An and Takao Terano, Tokyo Institute of Technology, Japan and Zhangbing Chen, Min Wu, Min Ding, and Hua Xie, China University of Geosciences, China

    Wind power prediction (WPP) is of great importance to the safety of the power grid and the effectiveness of power generations dispatching. However, the accuracy of WPP obtained by single numerical weather prediction (NWP) is difficult to satisfy the demands of the power system. In this research, we proposed a WPP method based on Bayesian fusion and multi-source NWPs. First, the statistic characteristics of the forecasted wind speed of each-source NWP was analysed, pre-processed and transformed. Then, a fusion method based on Bayesian method was designed to forecast the wind speed by using the multi-source NWPs, which is more accurate than any original forecasted wind speed of each-source NWP. Finally, the neural network method was employed to predict the wind power with the wind speed forecasted by Bayesian method. The experimental results demonstrate that the accuracy of the forecasted wind speed and wind power prediction is improved significantly.

  • Illumination Normalization in Images based on the Logarithm Transforms, Histogram Equalization and Discrete Cosine Transforms
    Rolando Jes'us C'ardenas Talavera, Edison Gregorio Paria Fern'andez and Juan Carlos Gutierrez Caceres, National University of San Agust'in Arequipa, Peru

    Illumination normalization is of utmost importance in several applications, there are researches showing several techniques proposed which have an advantages and disadvantages,but they are not clear when it comes to applying them. In this way, this work present and overview of the illumination normalization techniques, and three are chosen: the Logarithm Transform, the Histogram Equalization and the Discrete CosineTransform, these were compare with the Histogram Equalization function of Matlab, we obtained more correct restoration of image illumination than it, in addition, which are applied to vehicle and faces recognition, it can be seen in the results that the techniques improve recognition rates, which means they are of utmost importance in real applications.

  • Segmentation and Classification of Brain Tumor CT Images using SVM with Weighted Kernel Width
    Kimia Rezaei, Hamed Agahi, Islamic Azad University, Iran

    In this article a method is proposed for segmentation and classification of benign and malignant tumor slices inbrain Computed Tomography (CT) images. In this study image noises are removed using median and wiener filter and brain tumors are segmented using Support Vector Machine (SVM). Then a two-level discrete wavelet decomposition of tumor image is performed and the approximation at the second level is obtained to replace the original image to be used for texture analysis. Here, 17 features are extracted that 6 of them are selected using Student's t-test. Dominant gray level run length and gray level co-occurrence texture features are used for SVM training. Malignant and benign tumors are classified using SVM with kernel width and Weighted kernel width (WSVM) and k-Nearest Neighbors (k-NN) classifier. Classification accuracy of classifiers are evaluated using 10 fold cross validation method.Thesegmentation results are also compared with the experienced radiologist ground truth. The experimental results show that the proposed WSVM classifier is able to achieve high classification accuracy effectiveness as measured by sensitivity and specificity.

  • Human Activity Recognition Using Combined Deep Architectures
    Amsalu Tomas , K. K. Biswas, Indian Institute of Technology, India

    Many of computer vision problems now days are being solved by machine learning by learning representations from raw data. A good learning algorithm has to develop and learn internal representations of the visual object which are robust and invariant to variabilities by itself so that it can guess the features, factors or causes to represents the image. The best way to design such a learning algorithm so far is inspired by mammalian visual cortex, and called deep networks. Many recent researches have shown deep architectures are outperforming state-of-art in multidimensional fields, including human-activity recognition (HAR). The researcherused a combination of two of these deep architectures, Convolutional Neural Network (CNN) and Sparse Auto-encoders (SAE) on sampled image frames of videos to recognize human activities.
    The researcher propose a combined deep architecture, CNN to learn human action representations from Motion History Images (MHIs) of sampled image frames and SAE to learn discriminative movements of human skeletal joints on each sampled frame invariant to depth and size of the human object in the video clip. Both networks are trained separately, average the Softmax class posteriors across the sampled frames to obtain score of the clip, normalized class scores of each network in [0,1] and late fusion is performed by taking weighted mean of class scores of the networks.
    The proposed architecture is trained and evaluated on standard action recognition benchmarks of Microsoft-Research(MSR) Daily Activity3D and MSR Action3D datasets, where the proposed architecture has shown improvement on the state-of-art by 5% test accuracy.

  • Color Transform Based Approach for Detection and Classification of Plant Diseases
    Mosbah El Sghair, Mansur Jaba, Najib Tanish and Abdualsalam Aborgiga, John Naisbitt University, Lybia

    We propose and through an experiment valuate a software package resolution for automatic detection and classification of plant leaf diseases. The planned resolution is associate improvement to resolution planned in [1] because it provides quicker and additional correct solution. The developed process theme consists of several main phases as in [1]. the subsequent two steps are added in turn once the segmentation section. This is the primary and necessary section for automatic detection and classification of plant diseases. disease spots are totally different in color however not in intensity, compared with plant leaf color. thus we have a tendency to color remodel of RGB image is used for higher segmentation of disease spots. during this paper a comparison of the impact of CIELAB, HSI and YCbCr color area within the method of sickness spot detection is finished. Median filter is used for image smoothing. Finally threshold is calculated by applying Otsu methodology on color element to observe the disease spot. associate algorithmic program that is independent of ground noise, plant sort and disease spot color was developed and experiments were dole out on totally different family plant leaves with each, noise free (white) and clanging background.