基于MPSO-BP神经网络的煤矸图像识别方法研究

Research on image recognition method of coal gangue based on MPSO-BP neural network

为应对煤矸石图像对比度低、环境复杂、边界难以识别等问题,研究提出了一种基于改进型微粒群算法与神经网络的煤矸石识别方法。首先,对煤矸石图像进行预处理并提取纹理特征;其次,结合纹理特征与灰度均值,利用改进粒子群优化算法优化反向传播神经网络;最后,引入迁移学习与卷积神经网络模型。结果表明,结合灰度均值与纹理特征的神经网络在识别煤矸石时,比传统方法效果更佳,并且识别时间较短,仅为1980 s。引入迁移学习与卷积神经网络后,识别精确度在扩展数据库上分别提升了2.47%、1.47%和2.60%,改进后的模型性能精度高达0.95。与传统方法相比,研究方法在时间和识别精度上均有所提升,为煤炭图像质量的在线检测提供了理论与实践价值。

To deal with the problems of low contrast,complex environment and difficult boundary recognition of coal gangue images,a coal gangue recognition method based on improved particle swarm optimization algorithm and neural network was proposed.Firstly,the coal gangue image was preprocessed and the texture features were extracted.Secondly,the backpropagation neural network was optimized using improved particle swarm optimization algorithm by combining texture features and gray mean.Finally,transfer learning and convolutional neural network models were introduced.The results show that the neural network combined with gray mean and texture features is more effective than the traditional method in identifying coal gangue,and the recognition time is shorter,only 1980 s.After the introduction of transfer learning and convolutional neural network,the recognition accuracy of the extended database is improved by 2.47%,1.47%and 2.60%,respectively.The performance accuracy of the improved model is up to 0.95.Compared with the traditional methods,the research method is improved in both time and recognition accuracy,which provides theoretical and practical value for online coal image quality detection.