A quality of centroid-based clustering is highly dependent on initialization. In the article we propose initialization based on the probability of finding objects, which could represent individual clusters. We present...
详细信息
A quality of centroid-based clustering is highly dependent on initialization. In the article we propose initialization based on the probability of finding objects, which could represent individual clusters. We present results of experiments which compare the quality of clustering obtained by k-means algorithm and by selected methods for fuzzy clustering: FCM (fuzzy c-means), PCA (possibilistic clustering algorithm) and UPFC (unsupervised possibilistic fuzzy clustering) with different initializations. These experiments demonstrate an improvement in the quality of clustering when initialized by the proposed method. The concept how to estimate a ratio of added noise is also presented.
Soil roughness plays an essential role in the reflection of the incoming radar signal at the soil surface and is, therefore, highly important in the retrieval of the soil moisture information from the backscattered ra...
详细信息
Soil roughness plays an essential role in the reflection of the incoming radar signal at the soil surface and is, therefore, highly important in the retrieval of the soil moisture information from the backscattered radar signal. However, soil roughness, generally described by means of the root mean square (rms) height and the correlation length, remains difficult to measure correctly and is, furthermore, found to be highly variable. In order to overcome these difficulties, Verhoest et al. suggested the use of possibility distributions to reflect possible values of roughness parameters for a given roughness state of an agricultural field. These distributions were then further used to retrieve the soil moisture information. Nevertheless, as they estimated the possibility distributions by brute force, without taking into account any interactivity between the roughness parameters, rather wide distributions of retrieved soil moisture content were obtained. This paper first tries to independently estimate the possibility distributions for both roughness parameters on the basis of a synthetically generated roughness data set. Next, the interactivity between the rms height and the correlation length is taken into account through the identification of a joint possibility distribution by means of a possibilistic clustering algorithm. When applied to actual synthetic aperture radar data, the results show that a narrower, i.e., more specific, possibility distribution of the soil moisture content is obtained when the possibilistic retrieval procedure is performed based on the joint possibility distributions.
clustering is an unsupervised learning technique commonly used for image segmentation. As the outcome of most clusteringalgorithms is heavily dependent on the initial cluster centers, it is necessary to consider opti...
详细信息
ISBN:
(纸本)9781728101378
clustering is an unsupervised learning technique commonly used for image segmentation. As the outcome of most clusteringalgorithms is heavily dependent on the initial cluster centers, it is necessary to consider optimization during the process of segmentation. The Multi-Objective evolutionary algorithm (MOEA) was used for optimization in this study, to find optimal cluster centers. It is important to note that the effectiveness of MOEA is dependent upon the selection of objective functions. Two objectives were considered;namely, the minimization of intra-cluster compactness and the maximization of inter-cluster separation to determine the optimal initial cluster centers. Xie-Beni index (XBI) was used to measure the compactness and separation of cluster centers while the Average Inter-Cluster Separation (AIS) measure ensures the minimal overlapping of clusters. The MOEA will generate a set of non-dominated solutions. The Davies-Bouldin Index (DBI) is then employed to determine the most optimal solution for the cluster centers. Experimental results demonstrate that this method of image segmentation performs better than single-objective optimization (SOO)and possibilistic clustering algorithm (PCA).
暂无评论