基于加权中智C均值算法的变压器故障诊断模型

为提高变压器故障诊断准确度,提出了一种基于加权中智C均值算法的变压器故障诊断方法。该方法利用基于样本相似度的加权方法对样本特征进行加权,再引入中智理论对样本的分布重新分配,建立起基于加权中智C均值算法的变压器故障诊断模型。研究结果表明,该方法不仅弥补了传统FCM相同权重分配的不足,有效提高了故障诊断的准确率,且诊断结果产生的中智点对故障的变化预测具有重要意义。     

基于梯度提升决策树的车轮轮缘厚度磨耗预测

轮对在列车走行过程中起着导向、承受以及传递载荷的作用，其踏面及轮缘磨耗对地铁列车运行安全性和钢轨的寿命都将产生重要影响。根据地铁列车车轮磨耗机理，分析车轮尺寸数据特点，针对轮缘厚度这一型面参数，基于梯度提升决策树算法构建轮缘厚度磨耗预测模型。在该模型的基础上，任意选取某轮对数据进行验证分析，结果表明：基于梯度提升决策树的轮对磨耗预测模型具有较好的预测精度，可预测出1~6个月的轮缘厚度变化趋势范围，预测时间范围较长，可为地铁维保部门对轮对的维修方式由状态修转为预防修提供指导性建议。     

基于深度学习的秀丽隐杆线虫显微图像分割方法

利用计算机实现自动、准确的秀丽隐杆线虫(C.elegans)的各项形态学参数分析,至关重要的是从显微图像上分割出线虫体态,但由于显微镜下的图像噪声较多,线虫边缘像素与周围环境相似,而且线虫的体态具有鞭毛和其他附着物需要分离,多方面因素导致设计一个鲁棒性的C.elegans分割算法仍然面临着挑战。针对这些问题,提出了一种基于深度学习的线虫分割方法,通过训练掩模区域卷积神经网络(Mask R-CNN)学习线虫形态特征实现自动分割。首先,通过改进多级特征池化将高级语义特征与低级边缘特征融合,结合大幅度软最大损失(LMSL)损失算法改进损失计算;然后,改进非极大值抑制;最后,引入全连接融合分支等方法对分割结果进行进一步优化。实验结果表明,相比原始的Mask R-CNN,该方法平均精确率(AP)提升了4.3个百分点,平均交并比(mIOU)提升了4个百分点。表明所提出的深度学习分割方法能够有效提高分割准确率,在显微图像中更加精确地分割出线虫体。     

Supply chain data analytics for predicting supplier disruptions: a case study in complex asset manufacturing

Although predictive machine learning for supply chain data analytics has recently been reported as a significant area of investigation due to the rising popularity of the AI paradigm in industry, there is a distinct lack of case studies that showcase its application from a practical point of view. In this paper, we discuss the application of data analytics in predicting first tier supply chain disruptions using historical data available to an Original Equipment Manufacturer (OEM). Our methodology includes three phases: First, an exploratory phase is conducted to select and engineer potential features that can act as useful predictors of disruptions. This is followed by the development of a performance metric in alignment with the specific goals of the case study to rate successful methods. Third, an experimental design is created to systematically analyse the success rate of different algorithms, algorithmic parameters, on the selected feature space. Our results indicate that adding engineered features in the data, namely agility, outperforms other experiments leading to the final algorithm that can predict late orders with 80% accuracy. An additional contribution is the novel application of machine learning in predicting supply disruptions. Through the discussion and the development of the case study we hope to shed light on the development and application of data analytics techniques in the analysis of supply chain data. We conclude by highlighting the importance of domain knowledge for successfully engineering features.     

小尺度交通信号灯的检测与状态识别

交通信号灯识别包括检测和状态识别，在智能交通系统中发挥重要作用。基于YOLOv3算法提出了一种交通信号灯检测与状态识别模型。针对交通信号灯相较于交通场景中其他目标具有尺度小的特性进行了算法的设计：降低骨干网络的下采样倍率以增加小尺度目标的特征描述能力；通过增大特征图的尺度来改进多尺度特征融合；引入广义交并比作为检测任务的损失函数来改进目标边界框的回归效果。同时，根据交通信号灯本身的特性，使用颜色和形状约束的方法对信号灯进行状态识别和类别验证。最后在公开的Bosch交通信号灯数据集上和实际的城区道路进行了实验验证。实验结果表明，所提出的算法能够提升交通信号灯识别的精度和召回率，识别准确率可以达到90%左右。     

基于GFU和分层LSTM的组群行为识别研究方法

提出一种以"关键人物"为核心，使用门控融合单元（GFU，Gated Fusion Unit）进行特征融合的组群行为识别框架，旨在解决两个问题：①组群行为信息冗余，重点关注关键人物行为特征，忽略无关人员对组群行为的影响；②组群内部交互行为复杂，使用GFU有效融合以关键人物为核心的交互特征，再通过LSTM时序建模成为表征能力更强的组群特征.最终，通过softmax分类器进行组群行为类别分类.该算法在排球数据集上取得了86.7%的平均识别率.     

A self-adaptive weighted affinity propagation clustering for key frames extraction on human action recognition

In this paper, we propose a novel approach for key frames extraction on human action recognition from 3D video sequences. To represent human actions, an Energy Feature (EF), combining kinetic energy and potential energy, is extracted from 3D video sequences. A Self-adaptive Weighted Affinity Propagation (SWAP) algorithm is then proposed to extract the key frames. Finally, we employ SVM to recognize human actions on the EFs of selected key frames. The experiments show the information including whole action course can be effectively extracted by our method, and we obtain good recognition performance without losing classification accuracy. Moreover, the recognition speed is greatly improved.     

A New Optimized Wrapper Gene Selection Method for Breast Cancer Prediction

Machine-learning algorithms have been widely used in breast cancer diagnosis to help pathologists and physicians in the decision-making process. However, the high dimensionality of genetic data makes the classification process a challenging task. In this paper, we propose a new optimized wrapper gene selection method that is based on a nature-inspired algorithm (simulated annealing (SA)), which will help select the most informative genes for breast cancer prediction. These optimal genes will then be used to train the classifier to improve its accuracy and efficiency. Three supervised machine-learning algorithms, namely, the support vector machine, the decision tree, and the random forest were used to create the classifier models that will help to predict breast cancer. Two different experiments were conducted using three datasets: Gene expression (GE), deoxyribonucleic acid (DNA) methylation, and a combination of the two. Six measures were used to evaluate the performance of the proposed algorithm, which include the following: Accuracy, precision, recall, specificity, area under the curve (AUC), and execution time. The effectiveness of the proposed classifiers was evaluated through comprehensive experiments. The results demonstrated that our approach outperformed the conventional classifiers as expected in terms of accuracy and execution time. High accuracy values of 99.77%, 99.45%, and 99.45% have been achieved by SA-SVM for GE, DNA methylation, and the combined datasets, respectively. The execution time of the proposed approach was significantly reduced, in comparison to that of the traditional classifiers and the best execution time has been reached by SA-SVM, which was 0.02, 0.03, and 0.02 on GE, DNA methylation, and the combined datasets respectively. In regard to precision and specificity, SA-RF obtained the best result of 100 on GE dataset. While SA-SVM attained the best recall result of 100 on GE dataset.     

Hybrid Segmentation Scheme for Skin Features Extraction Using Dermoscopy Images

Objective and quantitative assessment of skin conditions is essential for cosmeceutical studies and research on skin aging and skin regeneration. Various handcraft-based image processing methods have been proposed to evaluate skin conditions objectively, but they have unavoidable disadvantages when used to analyze skin features accurately. This study proposes a hybrid segmentation scheme consisting of Deeplab v3+ with an Inception-ResNet-v2 backbone, LightGBM, and morphological processing (MP) to overcome the shortcomings of handcraft-based approaches. First, we apply Deeplab v3+ with an Inception-ResNet-v2 backbone for pixel segmentation of skin wrinkles and cells. Then, LightGBM and MP are used to enhance the pixel segmentation quality. Finally, we determine several skin features based on the results of wrinkle and cell segmentation. Our proposed segmentation scheme achieved a mean accuracy of 0.854, mean of intersection over union of 0.749, and mean boundary F1 score of 0.852, which achieved 1.1%, 6.7%, and 14.8% improvement over the panoptic-based semantic segmentation method, respectively.