自适应变步长高斯混合模型的工业烟尘目标分割方法

工业烟尘排放时的烟气黑度自动监测对提高环保质量和指导生产过程具有重要的应用价值, 针对传统的 高斯混合模型在进行背景建模时, 参数是在固定帧值的基础上进行参数更新, 导致烟尘检测不准确等问题, 提出一 种自适应变步长高斯混合模型的工业烟尘图像分割方法. 根据烟尘变化速度不均匀的特点, 通过分析检测出烟尘 与实际烟尘的检出率和检准率的和的最大值, 计算熵值差变化率对应的最佳步长, 得到熵值差变化率与最佳步长的 模型. 以熵值差变化率为依据, 确定最佳步长, 得到一个关于熵值差变化率与最佳步长的模型. 以熵值差变化率为 输入, 以最佳步长为输出, 在广义回归神经网络(GRNN)得到适用于本文工业烟尘图像分割的网络模型. 最后, 在多 个场景的烟尘视频中进行分割实验, 结果表明, 本文中方法能够有效的分割出视频中烟尘区域, 且具有一定的适用 性.

Method to segment industrial emission smoke target based on adaptive variable step Gauss mixture model

Automatic monitoring of smoke blackness during industrial smoke emission has important application value for improving environmental protection quality and guiding production process. Aiming at the problem of inaccurate smoke detection caused by updating parameters on the basis of previous fixed frame values in traditional Gauss mixture model for background modeling. A method to segment industrial emission smoke based on adaptive variable step is proposed. Combining the characteristics of uneven change speed of smoke, the maximum sum of detection rate and detection accuracy rate of smoke and dust is analyzed, and the optimal step length corresponding to the change rate of entropy difference is calculated, and the model of the change rate of entropy difference and the optimal step length is obtained. Based on the change rate of entropy difference, the optimum step size is determined, and a model about the optimum step size is obtained. With the change rate of entropy difference as input and the optimal step size as output, a network model is obtained in generalized regression neural network network. Finally, segmentation experiments are carried out in smoke several videos. The results shows the effectiveness of the proposed method.