针对双相机成像系统所获取的颗粒离焦图像，搭建了基于Faster-RCNN和VGG16卷积神经网络（CNN）的颗粒粒径与位置同步预测模型。在75~95 mm（约9~10倍的成像系统景深）深度范围内，拍摄直径为50~350 μm的9种不同圆点用于所提模型的训练，并将所提模型与基于离焦测距（DFD）理论模型的处理方法进行了比较分析。测量结果表明，与基于DFD理论模型的处理方法相比，CNN模型的颗粒深度测量范围有所提高，直径测量误差有所降低，但深度测量误差有所增大。进一步采用双相机系统拍摄在循环样品池内流动的粒径为120 μm的标准颗粒，应用所提CNN模型对相应图片进行处理，粒径预测结果的相对误差范围为-8%~8%。

High-order harmonic generation from one-dimensional (1D) multi-atom molecular ions in an ultra-short laser field is theoretically investigated. The dynamics of the electron in a linearly polarized intense laser field is analyzed in terms of 1D Schrodinger equation with the Crank-Nicolson algorithm. The dependence of high-order harmonics on the laser frequency and the inter-nuclear distance is discussed. It is found that the optimum range of inter-nuclear distance should be changed to get extended harmonic generation for different laser frequency, and the lower frequency laser pulse is favorable to higher order harmonic generation as the inter-nuclear distance increases.