基于神经网络的交互式炉膛火焰图像识别

炉膛火焰燃烧状态监测的关键技术之一是炉膛火焰图像的分类和识别。由于炉膛火焰燃烧过程的复杂性，使得准确反映炉膛火焰燃烧状态的火焰图像特征参数难以确定，在用神经网络训练方法构造分类器时，神经网络的收敛速度和识别的准确性不能同时满足实际要求。文中提出了交互式火焰图像识别方法,改善神经网络的分类识别性能。在神经网络的构造过程中，将人对神经网络分类器构造结果的评价信息反馈给网络，使其根据反馈信息进一步修正分类器。由于将人工的修正信息引入到分类器构造中，加快了神经网络的收敛速度，提高了神经网络识别的准确性。对4 000幅火焰图像的实验显示了此方法的有效性。

Interactive Method of Furnace Flame Image Recognition Based on Neural Networks

One of the key technologies used to monitor the furnace flame combustion condition is the classification and recognition of the furnace flame images. Owing to the complication of the furnace flame combustion process, it is difficult to determine which features of the flame image can accurately reflect the flame combustion condition. When the artificial neural network training approach is applied to construct classifiers, the convergence rate and the recognition accuracy of the neural networks can not meet the actual request at the same time. The interactive flame image recognition approach was introduced to improve the classification and recognition performance of the neural networks. In the processes of the neural networks construction, the human evaluation for the results of the neural network construction was returned to the neural networks to refine further the classifiers. Because the human adjustment was added in the classifiers' construction processes, the convergence rate of the neural networks is speeded up and the accuracy of the classification and recognition is enhanced. The performance of the approach was tested using 4 000 furnace flame images. Results demonstrate that the approach is competent.