Foreign fibers in cotton refer to non-cotton fibers and dyed fibers such as hairs, binding ropes, plastic films, candy wrappers, and polypropylene twines. Foreign fibers in cotton even in low content, especially in lint, can seriously affect the quality of the final cotton textile products. Today, online detection systems based on machine vision have been developed for evaluating the quality of the cotton. In such systems, classification of foreign fibers in cotton is the basic and key technology, which is related to the systems’ performance. Finding the optimum feature set with the small size and high accuracy is essential due to it can not only simplify the design of classifier, but also reduce the time of feature extraction. It is a feature selection problem in nature. Feature selection plays an important role in online detection of foreign fibers in cotton. This paper proposed a combined feature selection algorithm for foreign fiber data by combining Fisher Score with BPSO (Binary Particle Swarm Optimization). First, Fisher Score was used to filter noisy features. Then, the BPSO used the classifier accuracy as a fitness function to select the highly discriminating features. The proposed method was tested for classification on foreign fiber dataset. The comparisons of the proposed algorithm with Fisher Score approach and BPSO algorithm showed that the proposed algorithm was able to find the subsets with small size that produced the best classification accuracy in cross-validation. The optimal set with 18 features was selected from 75 features by the proposed algorithm, which classification accuracy reached 93.5%. The time cost of the optimal sets involving three stages corresponding to image segmentation, feature extraction and classification throughout the process of online detection was also tested. The time (0.8231 s) of the optimal set obtained by the proposed algorithm was obviously lower than the original set and the other subset selected by Fscore and BPSO. As a result, the optimal sets obtained by the proposed algorithm was more suitable to online detection and could effectively improve the performance of online detection systems.%为改进基于机器视觉的棉花异性纤维在线检测效率，提出一种基于费舍尔评分与离散粒子群优化的棉花异性纤维特征选择方法。该方法将费舍尔评分滤波式特征选择方法及基于离散粒子群优化的捆绑式特征选择方法组合在一起，首先利用费舍尔评分方法过滤噪声特征，然后利用离散粒子群算法从已去噪的特征集中选取最优特征子集。提出的方法应用于棉花异性纤维数据集，并与费舍尔评分方法、离散粒子群方法、遗传算法、蚁群算法进行对比，试验结果表明该方法可以更有效地选择出有较少特征数目、较高分类精度的特征子集。从75个棉花异性纤维原始特征中选出18个特征组成的特征集，其分类准确度达到93.5%，检测时间仅为0.8231 s，有效地改进了棉花异性纤维在线检测的精度与效率，从而减少异性纤维对棉纺织品的危害，提高棉纺企业经济效益。
展开▼
