搜索空间逐步缩小的遗传算法用于尿沉渣图像特征选择的研究

基于尿沉渣图像特征选择问题,提出一种新的特征优选方法,首先引入类空间分层分类思想,将多类成分特征集优选问题转化为两类成分的特征集优选问题以减少特征数,从而减少了后续分类器的维数复杂度并提高了优选后的特征集对成分的识别率;针对红细胞和白细胞的特征集优选问题,采用改进的遗传算法进行处理,先根据统计实验结果,锁定待选特征集中形态特征和纹理特征相应的两个可分度最大且相互独立的特征,然后使用基因位逐步锁定技术,结合小生境技术和自适应交叉变异算子,提高了遗传算法的搜索性能;最后,为了提高特征集的优选效果和稳定性,引入"多票投选"机制。就多个尿沉渣成分样本进行验证实验,结果表明,该算法优选的特征集与通过其他方式获得的特征集相比,识别率较高,而且明显减少了后续分类器的维数复杂度。

The Research of Feature Selection of Urinary Sediment Images Using Gradual Reduction of Searching Space for Genetic Algorithm

A new method was proposed based on urinary sediment image feature selection. Firstly, the idea of hierarchical classification within multiclass space was introduced, converting the feature selection about multi-class particles into the one about binary-class particles and reducing the number of the features and increasing the recognition rate; Then according to the feature selection issue about red cells and white cells, the modified genetic algorithm was adopted to deal with it. The modified genetic algorithm used gene-fixing technology and put the Niche technology and adaptive mutation operator together to enhance the performance of the genetic algorithm; Finally, in order to improve the effect and stability of the feature selection, voting mechanism was introduced. The test experiments were done, and the results of these experiments showed that this algorithm performed better than that of some other methods, the features were reduced and the complexity of the BPNN classifier was reduced as well.