CAJ | 학술논문

针对大量不相关的冗余特征制约图像关键特征分类性，导致语义识别模型出现偏差的问题，提出一种混合约束的半监督图像语义特征选择方法。在半监督学习的基础上，算法采用成对约束和数据清洗相结合的特征评价方法快速去除不相关图像干扰特征，聚类过程中考虑可转换语义直接的特征相关性，保证语义特征正确。实验结果表明，与传统的语义特征识别方法相比，该方法能以更少的特征获得更好的语义特征分类性能，有更好的识别效果。
침대대량불상관적용여특정제약도상관건특정분류성，도치어의식별모형출현편차적문제，제출일충혼합약속적반감독도상어의특정선택방법。재반감독학습적기출상，산법채용성대약속화수거청세상결합적특정평개방법쾌속거제불상관도상간우특정，취류과정중고필가전환어의직접적특정상관성，보증어의특정정학。실험결과표명，여전통적어의특정식별방법상비，해방법능이경소적특정획득경호적어의특정분류성능，유경호적식별효과。
Because the analysis of a large amount of irrelevant and redundant features restrict image key characteristics classification the semantic identification model appeared deviation,the paprer proposes a kind of hybrid constraint and semi-supervised image semantic feature selection method.Based on a semi-supervised learning the methods combined constraints in pairs and data cleaning to rapid removal irrelevant uses evaruation method which characteristics,clustring process consider convertible semantic direct correlation characteristics,ensure the correct semantic features.The experi-mentar results show that,with the traditionar semantic feature recognition method,this method can compared with less features get better semantic feature classification performance,have a better recognition effect.