摘要：In recent years, enteromorpha prolifera detectionhas received increasing attention. Supervised learning withremote sensing images can achieve satisfactory performancesfor green tide monitoring. However, data distributions betweenimages obviously differ, and it would be too costlyto label a massive amount of images for enteromorphaprolifera detection. Thus, this paper focuses on detectingenteromorpha prolifera using not only limited labelled data,but also a large amount of unlabelled data. We propose aneffective semi-supervised clustering framework for enteromorphaprolifera detection, which can reduce the labellingcost and alleviate the overfitting problem. Experimentalresults prove the effectiveness and potential of our approach,with almost a 15% increase from baseline. In addition, theproposed approach can provide quantitative assessments forband data of moderate resolution imaging spectroradiometer(MODIS) images, and several often ignored bands, such asbands 5, 6, and 7, are shown to be useful for enteromorphaprolifera detection.
