| 姬晓飞,石宇辰,王 昱,田晓欧.D-S 理论多分类器融合的光学遥感图像多目标识别[J].电子测量与仪器学报,2020,34(5):127-132 |
| D-S 理论多分类器融合的光学遥感图像多目标识别 |
| D-S theory based multi-classifier fusion optical remote sensing image target recognition |
| |
| DOI： |
| 中文关键词: 光学遥感图像 决策级融合 线性融合 D-S 证据理论 特征提取 多分类器 目标识别 |
| 英文关键词:optical remote sensing image decision fusion linear fusion D-S evidence theory feature extraction multi-classifier target recognition |
| 基金项目:国家自然科学基金(61906125)、辽宁省教育厅科学研究服务地方项目(L201708)、辽宁省教育厅科学研究青年项目(L201745)资助 |
|
|
|
|
| 摘要点击次数: 577 |
| 全文下载次数: 552 |
| 中文摘要: |
| 光学遥感图像的多目标检测与识别一直是图像处理与分析领域的热点研究问题。 针对多特征单一分类器决策级融合
不能很好的利用特征与分类器的适应性,导致识别的准确率很难进一步提高的问题,提出了基于 D-S 证据理论的多特征多分类
器决策级融合策略。 首先提取了两种简单且具有平移、缩放不变性的特征;其次分别引入 3 种适应性较好的分类器进行分类;
最后设计了两级的 D-S 证据理论的融合方案,并且在置信度函数计算的过程中引入表征分类器性能的混淆矩阵。 该算法有效
地解决了分类器输出的不确定性问题,进一步提高了光学遥感图像多目标分类识别的准确性。 测试表明,对 4 种目标的识别率
达到 97. 22%,验证了算法的有效性。 |
| 英文摘要: |
| The multi-target detection and recognition of optical remote sensing images have always been the hot researching topics in
image processing and analysis. The multi-target classification and recognition algorithm based on multiple features and single classifier
cannot make good use of the adaptability of features and classifiers, resulting in a problem that the accuracy of recognition is difficult to
improve. A multi-feature multi-classifier fusion optical target image recognition algorithm based on D-S evidence theory is proposed. Two
features with translation and scaling invariance are extracted. Secondly, three classifiers are introduced to classify the feature. Finally, a
two-level fusion algorithm scheme by using D-S evidence theory is proposed, and a confusion matrix that characterizes the performance of
the classifier is introduced in the calculation process of confidence function. The proposed algorithm is effectively resolved the classifier
output uncertainty problem, and further improves the accuracy of multi-target classification and recognition of optical remote sensing
images. The recognition rate of multi-objectives by DS evidence theory fusion strategy reaches 97. 22%. The effectiveness of the
algorithm is proved. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
