碩士 / 國立臺中科技大學 / 資訊工程系碩士班 / 106 / The main purpose of this study is to use the convolutional neural network (CNN) to carry on the automatic vessels detection and segmentation in the image. First, the objects of the vessels are detection from the image, and then these objects to the segmentation of the vessels type.
This paper uses the convoluted neural network architecture of YOLO, firstly, the vessels is detected and located from the image, and then the FCN and AirNet two convolutional neural network architecture as object segmentation model. And the position vessels is segmentation into the vessels object, and the optimized fused FCN and AirNet vessels segmentation results. And finally use the vessels segmentation results to find the main direction of the vessels.
In order to verify the effectiveness of this paper, this study carried out the detection and segmentation of vessels in this experiment. The two experimental images are select of the video from YouTube. The image data set is divided into three parts, The first part is training vessels detection, the data set has 144 images. The second part is the training vessels object segmentation, the data set has 1156 images. The third part is the test data set for vessels detection and vessels object segmentation, the data set has 549 test images. Experiments show that the detection results of this study, when the vessels is detected to have more than 80% of the area, the detection rate is 78.97%. Experiments show that the vessels object segmentation results with pixel accuracy (mPA) of up to 92%. Finally, in the calculation of the main direction of the vessels, the vessels main shaft direction can be used to guess the vessels running direction and track the suspicious vessels. From the detection and segmentation results, this paper shows that this method can effectively carried out the detection and segmentation of vessels. The future can be used in the vessels positioning and the vessels recognition.
Identifer | oai:union.ndltd.org:TW/106NTTI5392013 |
Date | January 2018 |
Creators | Wei-Ming Zeng, 曾偉銘 |
Contributors | 林春宏 |
Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
Language | zh-TW |
Detected Language | English |
Type | 學位論文 ; thesis |
Format | 57 |
Page generated in 0.0015 seconds