| 杜红彪,于伟,张旭,陈余庆.海面溢油图像特征识别双边分割算法研究[J].,2022,62(4):419-426 |
| 海面溢油图像特征识别双边分割算法研究 |
| Research on bilateral segmentation algorithms for feature recognition of sea surface oil spill images |
| |
| DOI：10.7511/dllgxb202204011 |
| 中文关键词: 溢油 图像识别 双边分割 双重注意力模块 |
| 英文关键词: oil spill imagere cognition bilateral segmentation dual attention module |
| 基金项目:国家自然科学基金青年基金资助项目(61203082)；大连海事大学研究生教育教学改革研究项目(YJG2020604). |
|
| 摘要点击次数: 349 |
| 全文下载次数: 454 |
| 中文摘要: |
| 重大海上溢油事故频发对海洋自然环境构成了巨大威胁．针对海面溢油图像的传统特征识别方法智能性、准确性不足等问题，探索了一种新型深度学习语义分割智能算法．首先分析了双边分割网络BiSeNetV2基本结构和功能模块单元．为了进一步降低现有网络参数复杂度，对其语义分支GE层进行了改进设计，提升了网络的轻量性．进而在BiSeNetV2的两个分支中引入双重注意力模块来解决类间相似性问题，增强了溢油图像特征识别的准确性．通过实验比较分析，验证了改进后的轻量型双边分割网络针对海面溢油图像特征识别准确率可达91.9%． |
| 英文摘要: |
| The frequent occurrence of major offshore oil spills poses a great threat to the marine natural environment. Aiming at the problems of insufficient intelligence and accuracy of the traditional feature recognition methods of sea surface oil spill images, a new intelligent algorithm of deep learning semantic segmentation is explored. Firstly, the basic structure and functional modules of bilateral segmentation network (BiSeNetV2) are analyzed. In order to further reduce the complexity of the existing network parameters, the GE layer of the semantic branch is improved to enhance the lightweightness of the network. Then, double attention module is added to the two branches of BiSeNetV2 to solve the problem of similarity between classes, which enhances the accuracy of oil spill image feature recognition. Through experimental comparison and analysis, it is verified that the recognition accuracy of the improved lightweight bilateral segmentation network for the characteristics of sea surface oil spill image can reach 91.9%. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|