Besides chemical composition, microstructure plays a key role to control the properties of engineering materials. A strong correlation exists between microstructure and many mechanical and physical properties of a metal. It has the utmost importance to understand the microstructure and distinguish the microstructure accurately for the appropriate selection of engineering materials in product fabrication. Computer vision and machine learning play a major role to extract the feature and predict the most probable class of a 7-class microstructural image with a high degree of accuracy. Features contain information about the image, and the classification function is defined in terms of features. Feature selection plays an important role in the classification problem to improve the classification accuracy and also to reduce the computational time by eliminating redundant or non-influential features. The current research aims at classifying microstructure image datasets by an improved wrapper-filter based feature selection method using texture-based feature descriptor. Before applying the feature selection method, a feature descriptor, called rotational local tetra pattern (RLTrP), is applied to extract the features from the input images. Then, an ensemble of three filter methods is developed by considering the union of the top-n features selected by Chi-square, Fisher score, and Gini impurity-based filter methods. The objective of this ensemble is to combine all possible important features selected by three filter methods which will be used to create an initial population of the wrapper-based meta-heuristic feature selection algorithm called, harmony search (HS). The novelty of this HS method lies in the objective function, which is defined as a function of Pearson correlation coefficient and mutual information to calculate the fitness value. The proposed method not only optimizes features with reduced dimension but also improves the performance of classification accuracy of the 7-class microstructural images. Moreover, the proposed HS model has also been compared with some standard optimization algorithms like whale optimization algorithm, particle swarm optimization, and Grey wolf optimization on the present dataset, and in every case, the HS method ensures better agreement between feature selection and classification accuracy than the other methods.

除化学成分外，微结构在控制工程材料的性能方面也起着关键作用。金属的微观结构与许多机械和物理特性之间存在很强的相关性。它具有最重要的是了解微观结构和准确区分微观结构，工程物资，在产品制造适当的选择。计算机视觉和机器学习在提取特征和高度准确地预测7类微结构图像中最可能的类中起着重要作用。特征包含有关图像的信息，并且分类功能是根据特征定义的。特征选择在分类问题中起着重要的作用，它通过消除冗余或无影响的特征来提高分类精度并减少计算时间。当前的研究旨在通过使用基于纹理的特征描述符的改进的基于包装滤波器的特征选择方法对微结构图像数据集进行分类。在应用特征选择方法之前，应用称为旋转局部四边形图案（RLTrP）的特征描述符从输入图像中提取特征。然后，通过考虑由卡方，费舍尔得分和基于基尼杂质的滤波方法选择的前n个特征的并集，开发出三种滤波方法的集合。该集合的目的是将通过三种过滤器方法选择的所有可能的重要特征进行组合，这三种过滤器方法将用于创建基于包装的基于元启发式特征选择算法（称为和声搜索（HS））的初始种群。该HS方法的新颖之处在于目标函数，该函数定义为Pearson相关系数和相互信息以计算适合度值的函数。所提出的方法不仅优化了尺寸减小的特征，而且提高了7类微结构图像的分类精度。此外，在现有数据集上，还将提出的HS模型与鲸鱼优化算法，粒子群优化和灰狼优化等一些标准优化算法进行了比较，在每种情况下，