提出了一种基于K近邻（K‑nearest neighbors，KNN）算法和相位敏感光时域反射（Phase‑sensitive optical time domain reflectometry，φ‑OTDR）系统的高铁声屏障故障识别方法。设计了V字型光缆敷设方式，能够感知声屏障不同高度吸声板在脉动力冲击下的振动，并利用φ‑OTDR系统采集振动信号。对振动信号进行多域特征提取以及K近邻分类后，可以实现对声屏障故障状态识别。实验结果表明，在复杂场景下对于故障点的识别正确率达到了90.9%。该方法为声屏障故障识别提供了一条可行的技术路线，能够减少对专业人员的依赖，对于提升高铁声屏障智能运维水平具有重要意义。

乳腺癌是世界上对于女性非常危险的疾病, 其患病率逐年增长, 是世界妇女死亡的主要原因。 在大样本情况下, 乳腺癌临床诊断受优质医疗资源相对短缺的限制, 诊断周期长、 检测费用高。 因此, 高效、 准确、 性价比高的乳腺癌诊断方法具有广阔的应用前景, 为临床诊断迫切需求。 荧光光谱检测是一种可以表征细胞中物理和化学综合变化的方法, 可用于表征正常和癌变细胞的特征。 机器学习擅长从大量数据中挖掘有用信息, 是进行分类和预测的有效手段。 以往机器学习多使用包含部分生化信息的数据库训练模型, 易导致信息缺失。 荧光光谱是细胞多种物质的叠加光谱, 使用荧光光谱特征峰诊断乳腺癌存在量化不确定性问题。 因此, 提出了机器学习结合乳腺癌样本荧光光谱的诊断方法。 使用405 nm波长的激光, 采集了正常和癌变乳腺组织(已做出病理诊断)的荧光光谱数据, 以此作为数据集, 比较了K-近邻(KNN)、 支持向量机(SVM)、 随机森林(RF)三种算法对正常和癌变乳腺组织荧光光谱的分类能力。 判别结果显示: 与SVM算法相比, KNN和RF算法的准确率更高、 平衡召回率和精度的能力更强, 对乳腺癌荧光光谱的分类能力更好, 其准确性、 召回率、 精度以及F1-score函数结果均在95%之上, 更利于乳腺癌的诊断。 进而探讨了权重KNN(WKNN)算法对正常和癌变乳腺组织荧光光谱的分类能力。 WKNN较KNN算法的分类评估结果有小幅度提升, 且具有更好的抗噪和适应能力, 算法简单。 综上所述, 本文提出的机器学习结合荧光光谱的乳腺癌诊断方法, 精度高、 速度快、 性价比高, 是未来乳腺癌诊断方法的发展方向, 具有重要的临床应用价值。

可溶性固形物含量（SSC）是反映苹果品质和成熟度的重要生理指标, 能够用于苹果品质分析和成熟度预测。 以新疆阿克苏冰糖心红富士苹果为研究对象, 从果实膨大定形期至完熟期, 以等间隔周期3 d采摘样本, 测其380~1 100 nm的可见/近红外光谱和SSC, 共552个样本。 然后融合分数阶微分（FD）及置换重要性-随机森林（PIMP-RF）算法, 构建成熟期苹果SSC预测的集成学习模型。 结果表明, 基于PLS模型优选的分数阶微分阶次为0阶、 0.4阶、 1.1阶和1.6阶, 且通过PIMP-RF算法进行特征重要性和可解释性分析结果显示, 利用可见/近红外光谱预测成熟期苹果SSC的关键波长主要为可见光波段, 这为今后研发新疆冰糖心红富士苹果的快速无损检测设备提供参考； 基于分数阶微分技术和PIMP-RF算法构建的成熟期苹果SSC集成学习模型具有很好的预测能力, 其训练集的相关系数r等于0.989 2, 平均绝对误差MAE等于0.241 2, 均方根误差RMSE等于0.309 1, 平均绝对百分误差等于0.018 3； 测试集的相关系数r等于0.903 8, 平均绝对误差MAE等于0.549 9, 均方根误差RMSE等于0.740 8, 平均绝对百分误差等于0.043 4, 相比于FD0-PIMP-RF、 FD0.4-PIMP-RF、 FD1.1-PIMP-RF和FD1.6-PIMP-RF模型, 集成学习模型为最优。 故而, 集成分数阶微分技术与PIMP-RF算法, 结合可见近红外光谱技术可有效地实现成熟期苹果的可溶性固形物含量预测。

强海杂波是对于对海雷达目标检测性能影响最大的干扰项, 目前已有的单一检测量和多特征联合检测算法性能极不稳定。针对上述难题, 提出基于聚类的对海雷达目标检测算法。所提算法提取了3个特征量--相对幅度方差(RAV)、相对平均能量(RAP)和相对向量熵(RVE), 对聚类算法中的k-近邻(kNN)算法进行模型改造, 完成了虚警可控的kNN检测器, 在该特征空间中利用kNN检测器实现目标与杂波分离。实测雷达数据实验结果显示, 在观测时间0.512 s与1.024 s时, 所提算法比基于分形的检测器的平均检测概率分别提高了56.2%和58.3%(HH极化模式下), 相比基于三特征的检测器的平均检测概率分别提高了29.2%和31.3%。可以得出结论: 所提算法能够实现复杂海况下的对海雷达目标检测, 且检测效果明显优于基于三特征的检测器算法。

针对图像匹配中AKAZE（Accelerated-KAZE）算法匹配精度较低以及计算复杂等问题，提出了一种基于高斯滤波和AKAZE-LATCH（AKAZE-Learned Arrangements of Three Patch Codes）算法相融合的图像匹配算法。首先，对输入图像进行高斯滤波预处理，去除高斯噪声等连续性噪声，并且保留图像的边缘信息。然后通过LATCH算法为AKAZE算法构建高效的二进制描述子，再通过KNN（K Nearest Neighbors）算法得到对应的匹配对。最后结合USAC(Universal RANSAC)去除误匹配对方法进行再次筛选，得到最终的匹配结果。经实验对比，所设计的算法相较于AKAZE算法匹配精度更高，且具有良好的鲁棒性和可靠性，可用于多数复杂场景下的图像匹配。

The paper reports results on the complex study on ferroelectric ceramics that represent solid solutions containing components with a perovskite-type or columbite-type structure. Solid solutions of a three-component (1−x−y)NaNbO3−xKNbO3−yCdNb₂O₆ system are manufactured at x = 0.05–0.20 and y = 0.10. Domain structures in ceramic grains are studied. The consistency between experimental and calculated results is examined for coexisting phases split into non-180∘ domains (mechanical twins) in the solid solution with x = 0.15. A correlation between the internal structure (crystal, domain, granular, and defect) and fundamental electromechanical and polarization properties is stated for the studied three-component solid solutions.

With the increasing demand of high-power and pulsed power electronic devices, environmental-friendly potassium sodium niobate ((Na0.5K0.5)NbO₃, KNN) ceramic-based capacitors have attracted much attention in recent years owning to the boosted energy storage density (Wrec). Nevertheless, the dielectric loss also increases as the external electric field increases, which will generate much dissipated energy and raise the temperature of ceramic capacitors. Thus, an effective strategy is proposed to enhance the energy storage efficiency (η) via tailoring relaxor behavior and bad gap energy in the ferroelectric 0.9(Na0.5K0.5)-NbO₃–0.1Bi(Zn2/3(NbxTa1−x)1/3)O₃ ceramics. On the one hand, the more diverse ions in the B-sites owing to introducing the Ta could further disturb the long-range ferroelectric polar order to form the short–range polar nanoregions (PNRs), resulting in the highη. On the other hand, the introduction of Ta ions could boost the intrinsic band energy gap and thus improve theEb. As a result, highWrec of 3.29 J/cm³ and ultrahighη of 90.1% at the high external electric field of 310 kV/cm are achieved inx = 0.5 sample. These results reveal that the KNN-based ceramics are promising lead-free candidate for high-power electronic devices.With the increasing demand of high-power and pulsed power electronic devices, environmental-friendly potassium sodium niobate ((Na0.5K0.5)NbO₃, KNN) ceramic-based capacitors have attracted much attention in recent years owning to the boosted energy storage density (Wrec). Nevertheless, the dielectric loss also increases as the external electric field increases, which will generate much dissipated energy and raise the temperature of ceramic capacitors. Thus, an effective strategy is proposed to enhance the energy storage efficiency (η) via tailoring relaxor behavior and bad gap energy in the ferroelectric 0.9(Na0.5K0.5)-NbO₃–0.1Bi(Zn2/3(NbxTa1−x)1/3)O₃ ceramics. On the one hand, the more diverse ions in the B-sites owing to introducing the Ta could further disturb the long-range ferroelectric polar order to form the short–range polar nanoregions (PNRs), resulting in the highη. On the other hand, the introduction of Ta ions could boost the intrinsic band energy gap and thus improve theEb. As a result, highWrec of 3.29 J/cm³ and ultrahighη of 90.1% at the high external electric field of 310 kV/cm are achieved inx = 0.5 sample. These results reveal that the KNN-based ceramics are promising lead-free candidate for high-power electronic devices.

在传统焦度计测量中, 由于被测镜片后表面曲率半径变化, 使得镜片后顶点与光阑表面无法完全重合, 造成焦度测量误差随镜片度数增大而增大的情况。针对此传统测量上无法克服的测量误差, 提出了一种利用不同焦度镜片图像提取不同特征的方法, 获得了不同焦度镜片的特征数据集, 并将特征数据集利用机器学习中k-近邻聚类算法(KNN algorithm)进行分类。选取折射率为1.551的-20 ~+20m-1每隔1m-1的单焦点镜片, 折射率为1.56与1.60的-15 ~+8m-1每隔0.5m-1的单焦点镜片进行镜片图像采集。实验结果表明, 所提方法下分类的精准度、召回率与F1评分均为100%, 三个模型均能正确识别其对应的所有测试镜片, 克服此测量误差。

This paper studies the dielectric spectra of solid solutions (SS) of the system (1 − x − y)NaNbO₃–xKNbO₃ –yCdNb₂O₆y= 0.075, x= 0.05 ÷ 0.30 in the temperature range (10 ÷ 900) K. The formation of a local maximum was established in the interval (260 ÷ 300) K at small x values, which, as KNbO₃ increases, is gradually blurred and becomes an inflection point. Detected in SS with x = 0.05 ÷ 0.10, the shift of the maxima of dependences 𝜀^′/ 𝜀₀(T) and 𝜀^′′/𝜀₀ (T), depending on the frequency of the electric field at the temperature ranges (300 ÷ 304) K and (258 ÷ 271) K, is not related to relaxation. This anomaly may indicate a crystallographic disorder to A and B positions. The conclusion is made about the expediency of using the obtained results for the development of functional ferroactive materials.This paper studies the dielectric spectra of solid solutions (SS) of the system (1 − x − y)NaNbO₃–xKNbO₃ –yCdNb₂O₆y= 0.075, x= 0.05 ÷ 0.30 in the temperature range (10 ÷ 900) K. The formation of a local maximum was established in the interval (260 ÷ 300) K at small x values, which, as KNbO₃ increases, is gradually blurred and becomes an inflection point. Detected in SS with x = 0.05 ÷ 0.10, the shift of the maxima of dependences 𝜀^′/ 𝜀₀(T) and 𝜀^′′/𝜀₀ (T), depending on the frequency of the electric field at the temperature ranges (300 ÷ 304) K and (258 ÷ 271) K, is not related to relaxation. This anomaly may indicate a crystallographic disorder to A and B positions. The conclusion is made about the expediency of using the obtained results for the development of functional ferroactive materials.

针对已有的三维圆形孔洞识别算法提取精度差、边缘点提取不完整等问题，提出一种基于点云法线与投影融合的三维圆孔识别的算法。首先利用KD-tree建立点云空间拓扑关系；其次使用近邻算法（KNN）搜索距离点最近的k个邻域点，通过定义距离阈值确定大于阈值的点为边界点；最后采用点云法线与投影融合，实现点云边缘的特征点和噪声点区分，并提取点云数据的三维圆孔特征。以带有圆形孔洞的栅极件为试验工件进行算法验证，试验结果表明，该算法能够有效实现点云边缘提取以及三维圆孔识别。