Bubble Plume Target Detection Method of Multibeam Water Column Images Based on Bags of Visual Word Features

Bubble plumes, as main manifestations of seabed gas leakage, play an important role in the exploration natural hydrate and other resources. Multibeam water column images have been widely used detecting bubble plume targets recent years because they can wholly record backscatter strengths. However, strong noises multibeam cause many issues target detection, traditional detection methods are mainly optical less efficient for noise-affected sonar images. To improve accuracy images, this study proposes a method based on bag visual words (BOVW) features support vector machine (SVM) classifier. First, characteristics analyzed, with conclusion that BOVW well express gray scale, texture, shape plumes. Second, constructed following steps point description extraction, clustering, feature encoding. Third, quadratic SVM classifier is recognition Finally, procedure described. In experiment using measured data Strait Georgia, proposed achieved 98.6% validation sets, 91.7% correct rate By comparison methods, experimental results prove validity method, show potential applications our research ocean