结合SVM的交互式遗传算法及其应用

Improved Interactive Genetic Algorithm Incorporating with SVM and Its Application

交互式遗传算法在人机交互过程中 ,用户对每代的所有个体都要进行评估。针对个体数量较大 ,进化代数较多 ,用户容易疲劳这一问题 ,提出了一种改进算法 ,充分利用遗传初始阶段用户所选择的正例和反例信息 ,采用支持向量机构造分类器 ,在整个图像库中找出更多符合分类器的图像加入到遗传过程 ,以扩大遗传操作中好个体的个数 ,加速算法收敛 ,从而减轻用户疲劳 ;同时交互中不断扩大的样本集也使支持向量机分类器更加准确 ,从而建立比较完善的个性化的情感模型。本文将该算法应用于服装图像的个性化情感检索。实验结果表明 ,所提出的方法可以较好地减轻用户疲劳 ,检索出的图像较好地体现用户的个性化情感。

User fatigue is a key technical problem to interactive genetic algorithm (IGA), since users should evaluate all individuals in every generation during interaction. Especially in the case that the number of individuals and generations become large, users will feel tired. An improved IGA incorporating with SVM is proposed, which fully mine the information in good individuals and bad ones at initial stage of IGA. A support vector machine is used to construct a classification by learning from these positive and negative samples and to classify all images in the database into two classes: positive and negative classes. Then the individuals in the GA population with lower fitness are replaced with the some best images in the positive class. Thus the number of good individuals in each generation is increased, the acceleration of GA convergence is expected. Meanwhile, SVM also constructs an individual emotional model by learning. The algorithm is used in individual emotion fashion image retrieval system. Experimental results demonstrate that the algorithm can alleviate user fatigue and has a good performance in individual emotional image retrieval.