SO₂紫外相机因在时间分辨率、空间分辨率、探测灵敏度以及探测精度等诸多方面均具有显著优势而成功应用于火山活动监测及其动力学研究。为解决紫外相机反演SO₂排放速率容易受烟羽湍流及图像低对比度影响等问题，提出了融入神经网络的光流算法。首先，基于大气紫外辐射传输特性，阐述了SO₂紫外相机的工作机理及SO₂浓度图像的反演方法；其次，将神经网络融入光流算法，实现了火山烟羽图像中SO₂排放速率的精确反演；最后，与传统光流法进行对比，论证了神经网络光流算法的科学性及优越性与精确性。实验结果表明：在图像低对比度及烟羽湍流效应的双重影响下，神经网络光流法可以把边缘反演的误差从94%降低至5%，显著提高了SO₂排放速率反演的精确性。

SO₂ UV camera has been successfully applied in volcanic activity monitoring and its dynamics research due to its remarkable advantages in temporal resolution, spatial resolution, detection sensitivity, and detection accuracy. To address the issues that SO₂ emission rate retrieved from UV camera images is easily affected by plume turbulence and the imges obtined are often with low contras, an optical flow algorithm incorporating neural network is proposed in this work. Firstly, based on the characteristics of atmospheric ultraviolet radiation transmission, the working mechanism of the SO₂ UV camera and the inversion method of SO₂ concentration image are described. Secondly, the neural network is integrated into the optical flow algorithm to achieve accurate inversion of SO₂ emission rate in volcanic plume images; Finally, compared with the traditional optical flow methods, the superiority and accuracy of the proposed neural network optical flow algorithm is confirmed. The experimental results show that the neural network optical flow method can reduce the error of edge inversion from 94% to 5% even under the dual influence of low contrast of images and plume turbulence effect, significantly improving the accuracy of SO₂ emission rate inversion.