ID	原文	译文
2443	相比最优非线性处理，该方法无需噪声参数估计，因而更具实用性。	Due to no parameter estimation, this method is more practical than the optimal nonlinearities.
2444	为满足多品种小批次、大规模定制模式下有效划分产品族的需求，全面分析 BOM(Bill of Materials，物料清单)所包含的特征，概括已有结构近似方法并提出内容近似度量模型，在此基础上提出组合两者的集成模型。	In order to meet the requirements of grouping product families for advanced manufacturing modes such asmass customization, the features in BOM (Bill of Materials)are comprehensively analyzed, and a concept of BOM struc-ture-based similarity metric model, a content-based similarity metric model, and an ensemble model combined with both are proposed.
2445	结构近似模型方面，以包含 BOM 层次结构和物料数量的相邻矩阵表示 BOM，利用正交普氏分析法计算 BOM 与 BOM 之间的近似程度。	In the structure-based model, BOMs are represented by adjacent matrixes, including the relationships between ma-terials and the quantity of materials, and the Orthogonal Procrustes Analysis is implemented to measure the similarity amongBOMs.
2446	内容近似模型方面，从 BOM 文本中提取有效特征，引入逆向词频法将文本特征转换成机器可识别向量形式，采用余弦近似公式完成向量近似的计算。	While in content-based model, effective text features are extracted from BOMs, being transformed to vectors byTFIDF(Term Frequency-Inverse Document Frequency), and finally being inputted into cosine approximation formula forsimilarity value.
2447	集成模型提出基于基尼系数的权重分配方法集成结构和内容两种模型。	To obtain more accuracy and performance, a weight distribution method based on the Gini coefficient is pro-posed for the ensemble model.
2448	最后，提供测试框架并通过实验评价集成模型较已有方法在模型性能及训练耗时上的优劣。	Finally, a test framework is provided and all models are in evaluated experimentally in accura-cy and performance.
2449	针对 RGB 图像具有丰富的色彩细节特征，红外图像对目标轮廓、尺寸、边界等外形特征有较高敏感度的特点，提出了一种非对称并行语义分割模型 APFCN(Asymmetric Parallelism Fully Convolutional Networks)。	Aiming at that RGB image is rich in color details of scene and infrared image is sensitive to outline、sizeand boundary of target, a novel semantic segmentation model APFCN (Asymmetric Parallelism Fully Convolutional Net-works)is proposed.
2450	APFCN 上路设计了一个卷积核尺寸非统一的五层空洞卷积网络来提取红外图像目标高层轮廓特征;	In the upper part of APFCN, a five layer dilation convolution network, where the five kernel sizes arenot uniform, is designed used to extract the high-level targets contour features of infrared image.
2451	下路沿用卷积加池化网络提取 RGB 图像三个尺度上的细节特征;	In the lower part of APF-CN, a classical CNN network is used to extract three scale features of RGB images.
2452	后端将红外图像高层特征与 RGB 图像三个尺度的细节特征进行融合，并将 4倍上采样后的融合特征作为语义分割输出。	At the back of APFCN, the high levelfeatures of the infrared image are fused with the three scale features of the RGB image, and the fused features after 4 timesupper sampling is used as the semantic segmentation output of APFCN.