混合优化RBF-BP网络的板形缺陷识别

针对传统板形模式识别方法存在精度低、鲁棒性弱的问题,提出了一种混合优化RBF-BP组合神经网络板形模式识别方法。首先利用自组织映射网络(SOM)对样本聚类,利用聚类后的网络拓扑结构确定RBF的中心,并计算RBF的宽度,克服了传统聚类算法随机选取中心导致聚类结果不稳定的问题。然后利用遗传算法(GA)良好的全局搜索能力优化整个网络的权值。RBF-BP组合神经网络是由一个RBF子网和一BP子网串联构成的,该网络同时具备BP神经网络能较好地预测未知样本的能力以及RBF神经网络的逼近速度快的优点。并以某900HC可逆冷轧机板形识别为应用背景,在MATLAB2010a环境下进行仿真实验,结果表明混合优化RBF-BP组合神经网络的板形模式识别方法能够识别出常见的板形缺陷,提高了板形缺陷识别精度并具有较好的鲁棒性,可以满足板带轧机高精度的板形控制要求。     

GA优化T-S云推理网络火灾识别模型设计

针对传统火灾报警系统存在着准确度不高、误报、漏报及泛化能力不强的问题,设计了一种基于GA(Genetic Algorithm)优化T-S云推理网络火灾探测模型,对模型进行训练和测试。并将T-S云推理网络与模糊神经网络对火灾信号的识别结果进行对比,给出MATLAB的仿真结果。通过仿真结果表明,该火灾探测模型识别精度更高,偏离样本数据测试结果与期望输出之间误差更小,模型提高了对火灾信号的识别精度和泛化能力。     

Spatially modulated scene illumination for intensity-compensated two-dimensional array photon-counting LiDAR imaging

Photon-counting LiDAR using a two-dimensional (2D) array detector has the advantages of high lateral resolution and fast acquisition speed. The non-uniform intensity profile of the illumination beam and non-uniform quantum efficiency of the detectors in the 2D array deteriorate the imaging quality. Herein, we propose a photon-counting LiDAR system that uses a spatial light modulator to control the spatial intensity to compensate for both the non-uniform intensity profile of the illumination beam, and the variation in the quantum efficiency of the detectors in the 2D array. By using a 635 nm peak wavelength and 4 mW average power semiconductor laser, lab-based experiments at a 4.27 m stand-off distance are performed to verify the effectiveness of the proposed method. Compared with the unmodulated method, the standard deviation of the intensity image of the proposed method is reduced from 0.109 to 0.089 for a whiteboard target, with an average signal photon number of 0.006 per pixel.