Microscopic images of bovine milk somatic cells are used to classify neutrophils, epithelial cells, macrophages and lymphocytes. Using pattern recognition technology, the problem of classification and recognition is solved from different nature, levels and spaces. The proposed RKSGA-SVM algorithm is used to realize somatic cell image recognition. First, color, morphological and texture features of four types of cells are extracted separately, including geometric and moment invariant features. Second, ReliefF algorithm is used to calculate the weights of all features. According to preset cumulative contribution rate, the preliminary feature set is obtained. Third, redundant features are eliminated by Kolmogorov–Smirnov (KS) test, and the high-level optimization is obtained. The selected feature sets have remarkable distinguishing ability. Finally, the weighted optimal feature sets are obtained by weighted coefficient method on the advanced optimal feature sets. The overall accuracy of RKSGA-SVM algorithm is 99.00%, and Kappa coefficient is 0.987. The proposed algorithm has the advantages of balancing classification accuracy, eliminating redundancy and reducing feature dimension. On the premise of ensuring high classification accuracy, the feature set can reduce feature dimension and the amount of data calculation, improve operation efficiency and save storage space. Experiments show that the feature selection method proposed in this paper is feasible and more suitable for extracting feature sets in the process of somatic cell classification.

中文翻译：

---

基于特征融合的多形牛体细胞识别

牛乳体细胞的显微图像用于对中性粒细胞、上皮细胞、巨噬细胞和淋巴细胞进行分类。利用模式识别技术，从不同的性质、层次和空间上解决分类识别问题。提出的RKSGA-SVM算法用于实现体细胞图像识别。首先，分别提取四种细胞的颜色、形态和纹理特征，包括几何特征和矩不变特征。其次，ReliefF算法用于计算所有特征的权重。根据预设的累积贡献率，得到初步的特征集。第三，通过 Kolmogorov-Smirnov (KS) 检验消除冗余特征，获得高级优化。所选特征集具有显着的区分能力。最后，加权最优特征集是在高级最优特征集上通过加权系数法得到的。RKSGA-SVM算法整体准确率为99.00%，Kappa系数为0.987。该算法具有平衡分类精度、消除冗余、降低特征维数等优点。该特征集在保证高分类精度的前提下，可以降低特征维数和数据计算量，提高运算效率，节省存储空间。实验表明，本文提出的特征选择方法是可行的，更适用于体细胞分类过程中的特征集提取。RKSGA-SVM算法整体准确率为99.00%，Kappa系数为0.987。该算法具有平衡分类精度、消除冗余、降低特征维数等优点。该特征集在保证高分类精度的前提下，可以降低特征维数和数据计算量，提高运算效率，节省存储空间。实验表明，本文提出的特征选择方法是可行的，更适用于体细胞分类过程中的特征集提取。RKSGA-SVM算法整体准确率为99.00%，Kappa系数为0.987。该算法具有平衡分类精度、消除冗余、降低特征维数等优点。该特征集在保证高分类精度的前提下，可以降低特征维数和数据计算量，提高运算效率，节省存储空间。实验表明，本文提出的特征选择方法是可行的，更适用于体细胞分类过程中的特征集提取。该特征集可以降低特征维数和数据计算量，提高运算效率，节省存储空间。实验表明，本文提出的特征选择方法是可行的，更适用于体细胞分类过程中的特征集提取。该特征集可以降低特征维数和数据计算量，提高运算效率，节省存储空间。实验表明，本文提出的特征选择方法是可行的，更适用于体细胞分类过程中的特征集提取。