The Research And Application On Classification In SAR Oil Spill Images Based On Agglomerative Clustering

In recent years, rapid economic development has led to sharp increase in oil consumption, oil imports and a large-scale mining let the ocean offshore oil major oil spill probability spiking upward. Synthetic Aperture Radar (Synthetic Aperture Radar, referred SAR) due to work with all-time characteristics of the oil spill detection,played a significant role. By sea, ocean currents affect the oil spill area shape characteristic change very quickly, so how to efficiently carry out oil spill area classification has important research and engineering value.SAR is a reflected microwave signal imaging apparatus, this imaging mechanism causes the image generally has serious speckle noise, and uneven striping equipment intensity distribution characteristics, which brings inconvenience for subsequent ientif-ication.This article study oil spill in SAR image classification based on region growing. Seeded point based on the traditional region growing algorithm execution is slow, and the classification results are heavily dependent on the initial seed selection, thus limiting its engineering application value. Watershed segmentation algorithm performs faster and divide image with boundary which is a single pixel wide,closed commun-icating precise contours. However, watershed segmentation algorithm applied directly to the oil spill SAR image’s biggest drawback is prone to over-segmentation. This article first use watershed segmentation algorithm for generating a plurality of regions of the image regional expression, thereby screening seed region, and then use the proposed fusion rule to achieve regional growth, namely clustering fusion of SAR oil spill image classification. Through the analysis of a large number of experimental results, the proposed fusion algorithm performs clustering method based on seed region is faster than the region growing algorithm, and acrquire more accuracy of oil spills on SAR image classification. The main research work carried out from the following three aspects:(1) SAR massive oil spill image to the target based. SRAD (anisotropic filtering) can effectively remove oil spill SAR image speckle noise, but also to maintain a good edge and target characteristics, and thus, to some extent inhibited the formation of over-segmentation of watershed segmentation phenomenon.(2) In this paper, based on the control marker watershed segmentation algorithm can further suppress the over-segmentation phenomenon. Reduce initial segmentation regions can greatly enhance regional growth and growth rate.(3) Based on the watershed algorithm based on clustering method used to achieve the integration of regional growth, according to the area of each region using different distance metric weights than the small area for an effective integration, to achieve effective oil spill SAR image classification.