• Proceedings of the Korean Society for Agricultural Machinery Conference
• /
• 2017.04a
• /
• pp.42-42
• /
• 2017

Prediction of rice yield during a growing season would be very helpful to magnify rice yield as it also allows better farm practices to maximize yield with greater profit and lesser costs. UAV imagery based automatic detection of rice can be a relevant solution for early prediction of yield. So, we propose an image processing technique to predict rice yield using low altitude UAV images. We proposed $L^*a^*b^*$ color space based image segmentation algorithm. All images were captured using UAV mounted RGB camera. The proposed algorithm was developed to find out rice grain area from the image background. We took RGB image and applied filter to remove noise and converted RGB image to $L^*a^*b^*$ color space. All color information contain in both $a^*$ and $b^*$ layers and by using k-mean clustering classification of these colors were executed. Variation between two colors can be measured and labelling of pixels was completed by cluster index. Image was finally segmented using color. The proposed method showed that rice grain could be segmented and we can recognize rice grains from the UAV images. We can analyze grain areas and by estimating area and volume we could predict rice yield.

• Journal of KIISE:Software and Applications
• /
• v.35 no.4
• /
• pp.255-264
• /
• 2008

This paper proposes a novel method using K-means and Non-negative matrix factorization (NMF) for topic -based multi-document summarization. NMF decomposes weighted term by sentence matrix into two sparse non-negative matrices: semantic feature matrix and semantic variable matrix. Obtained semantic features are comprehensible intuitively. Weighted similarity between topic and semantic features can prevent meaningless sentences that are similar to a topic from being selected. K-means clustering removes noises from sentences so that biased semantics of documents are not reflected to summaries. Besides, coherence of document summaries can be enhanced by arranging selected sentences in the order of their ranks. The experimental results show that the proposed method achieves better performance than other methods.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.4
• /
• pp.1535-1554
• /
• 2016

This paper proposes a novel channel assignment scheme called multi-cluster based dynamic channel assignment (MC-DCA) to improve system performance for the downlink of dense femtocell networks (DFNs) based on orthogonal frequency division multiple access (OFDMA) and frequency division duplexing (FDD). In order to dynamically assign channels for femtocell access points (FAPs), the MC-DCA scheme uses a heuristic method that consists of two steps: one is a multiple cluster assignment step to group FAPs using graph coloring algorithm with some extensions, while the other is a dynamic subchannel assignment step to allocate subchannels for maximizing the system capacity. Through simulations, we first find optimum parameters of the multiple FAP clustering to maximize the system capacity and then evaluate system performance in terms of the mean FAP capacity, unsatisfied femtocell user equipment (FUE) probability, and mean FAP power consumption for data transmission based on a given FUE traffic load. As a result, the MC-DCA scheme outperforms other schemes in two different DFN environments for commercial and office buildings.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2021.11a
• /
• pp.681-684
• /
• 2021

In this paper, we study image clustering without labeling using machine learning techniques. We proposed an unsupervised machine learning technique to design an image clustering model that automatically categorizes images into groups. Our experiment focused on inception convolutional neural networks (inception V3) with k-mean methods to cluster images. For this, we collect the public datasets containing Food-K5, Flowers, Handwritten Digit, Cats-dogs, and our dataset Rice Germination, and the owner dataset Palm print. Our experiment can expand into three-part; First, format all the images to un-label and move to whole datasets. Second, load dataset into the inception V3 extraction image features and transferred to the k-mean cluster group hold on six classes. Lastly, evaluate modeling accuracy using the confusion matrix base on precision, recall, F1 to analyze. In this our methods, we can get the results as 1) Handwritten Digit (precision = 1.000, recall = 1.000, F1 = 1.00), 2) Food-K5 (precision = 0.975, recall = 0.945, F1 = 0.96), 3) Palm print (precision = 1.000, recall = 0.999, F1 = 1.00), 4) Cats-dogs (precision = 0.997, recall = 0.475, F1 = 0.64), 5) Flowers (precision = 0.610, recall = 0.982, F1 = 0.75), and our dataset 6) Rice Germination (precision = 0.997, recall = 0.943, F1 = 0.97). Our experiment showed that modeling could get an accuracy rate of 0.8908; the outcomes state that the proposed model is strongest enough to differentiate the different images and classify them into clusters.

• Journal of Electrical Engineering and Technology
• /
• v.13 no.6
• /
• pp.2511-2520
• /
• 2018

In this study, Multi-Radial Basis Function Support Vector Machine (Multi-RBF SVM) classifier is introduced based on a composite kernel function. In the proposed multi-RBF support vector machine classifier, the input space is divided into several local subsets considered for extremely nonlinear classification tasks. Each local subset is expressed as nonlinear classification subspace and mapped into feature space by using kernel function. The composite kernel function employs the dual RBF structure. By capturing the nonlinear distribution knowledge of local subsets, the training data is mapped into higher feature space, then Multi-SVM classifier is realized by using the composite kernel function through optimization procedure similar to conventional SVM classifier. The original training data set is partitioned by using some unsupervised learning methods such as clustering methods. In this study, three types of clustering method are considered such as Affinity propagation (AP), Hard C-Mean (HCM) and Iterative Self-Organizing Data Analysis Technique Algorithm (ISODATA). Experimental results on benchmark machine learning datasets show that the proposed method improves the classification performance efficiently.

• Journal of information and communication convergence engineering
• /
• v.17 no.4
• /
• pp.261-266
• /
• 2019

Differential evolution (DE) is an uncomplicated and serviceable developmental algorithm. Nevertheless, its execution depends on strategies and regulating structures. The combination of several strategies between subpopulations helps to stabilize the probing on DE. In this paper, we propose a unique k-mean soft island model DE(KSDE) algorithm which maintains population diversity through soft island model (SIM). A combination of various approaches, called KSDE, intended for migrating the subpopulation information through SIM is developed in this study. First, the population is divided into k subpopulations using the k-means clustering algorithm. Second, the mutation pattern is singled randomly from a strategy pool. Third, the subpopulation information is migrated using SIM. The performance of KSDE was analyzed using 13 benchmark indices and compared with those of high-technology DE variants. The results demonstrate the efficiency and suitability of the KSDE system, and confirm that KSDE is a cost-effective algorithm compared with four other DE algorithms.

• Journal of Intelligence and Information Systems
• /
• v.7 no.2
• /
• pp.35-49
• /
• 2001

Over the years, cluster analysis has become a popular tool for marketing and segmentation researchers. There are various methods for cluster analysis. Among them, K-means partitioning cluster analysis is the most popular segmentation method. However, because the cluster analysis is very sensitive to the initial configurations of the data set at hand, it becomes an important issue to select an appropriate starting configuration that is comparable with the clustering of the whole data so as to improve the reliability of the clustering results. Many programs for K-mean cluster analysis employ various methods to choose the initial seeds and compute the centroids of clusters. In this paper, we suggest a methodology to evaluate various clustering programs. Furthermore, to explore the usability of the methodology, we evaluate four clustering programs by using the methodology.

• Journal of Environmental Science International
• /
• v.15 no.12
• /
• pp.1125-1139
• /
• 2006

The cluster analysis of diurnal precipitation patterns is performed by using daily precipitation of 59 stations in South Korea from 1973 to 1996 in four seasons of each year. Four seasons are shifted forward by 15 days compared to the general ones. Number of clusters are 15 in winter, 16 in spring and autumn, and 26 in summer, respectively. One of the classes is the totally dry day in each season, indicating that precipitation is never observed at any station. This is treated separately in this study. Distribution of the days among the clusters is rather uneven with rather low area-mean precipitation occurring most frequently. These 4 (seasons)$\times$2 (wet and dry days) classes represent more than the half (59 %) of all days of the year. On the other hand, even the smallest seasonal clusters show at least $5\sim9$ members in the 24 years (1973-1996) period of classification. The cluster analysis is directly performed for the major $5\sim8$ non-correlated coefficients of the diurnal precipitation patterns obtained by factor analysis In order to consider the spatial correlation. More specifically, hierarchical clustering based on Euclidean distance and Ward's method of agglomeration is applied. The relative variance explained by the clustering is as high as average (63%) with better capability in spring (66%) and winter (69 %), but lower than average in autumn (60%) and summer (59%). Through applying weighted relative variances, i.e. dividing the squared deviations by the cluster averages, we obtain even better values, i.e 78 % in average, compared to the same index without clustering. This means that the highest variance remains in the clusters with more precipitation. Besides all statistics necessary for the validation of the final classification, 4 cluster centers are mapped for each season to illustrate the range of typical extremities, paired according to their area mean precipitation or negative pattern correlation. Possible alternatives of the performed classification and reasons for their rejection are also discussed with inclusion of a wide spectrum of recommended applications.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.6 no.9
• /
• pp.2191-2201
• /
• 2012

Traffic classification seeks to assign packet flows to an appropriate quality of service (QoS) class based on flow statistics without the need to examine packet payloads. Classification proceeds in two steps. Classification rules are first built by analyzing traffic traces, and then the classification rules are evaluated using test data. In this paper, we use self-organizing map and K-means clustering as unsupervised machine learning methods to identify the inherent classes in traffic traces. Three clusters were discovered, corresponding to transactional, bulk data transfer, and interactive applications. The K-nearest neighbor classifier was found to be highly accurate for the traffic data and significantly better compared to a minimum mean distance classifier.

• Journal of the Korean Society of Manufacturing Technology Engineers
• /
• v.19 no.3
• /
• pp.326-333
• /
• 2010

In this study, DAQ and TRA modules were applied to the CFRP single specimen testing method using AE. A method for crack identification in CFRP specimens based on k-mean clustering and wavelet transform analysis are presented. Mode I on DCB under vertical loading and mode II on 3-points ENF testing under share loading have been carried out, thereafter k-mean method for clustering AE data and wavelet transition method per amplitude have been applied to investigate characteristics of interfacial fracture in CFRP composite. It was found that the fracture mechanism of Carbon/Epoxy Composite to estimate of different type of fractures such as matrix(epoxy resin) cracking, delamination and fiber breakage same as AE amplitude distribution using a AE frequency analysis. In conclusion, the presented results provide a foundation for using wavelet analysis as efficient crack detection tool. The advantage of using wavelet analysis is that local features in a displacement response signal can be identified with a desired resolution, provided that the response signal to be analyzed picks up the perturbations caused by the presence of the crack.