CHN Energy Shendong Coal Group Coal Preparation Center
School of Chemistry and Environmental Engineering, China University of Mining and Technology-Beijing

为提高煤与矸石分选的自动化与智能化程度,针对煤与矸石在线识别的过程中,图像特征值需人工选取且模型鲁棒性差的问题,以现场采集的煤与矸石原始图像作为输入,建立了一种基于卷积神经网络的煤与矸石图像识别模型。通过反卷积对卷积神经网络进行可视化处理,分析了卷积神经网络提取煤与矸石图像特征的过程,并以此为基础在卷积神经网络中设置小波变换层,利用Biorthogonal小波对原始图像进行分解,将高频系数与原始图像结合后进行卷积操作,优化了模型的识别效果。结果表明:该识别模型能够对煤与矸石图像进行有效识别,设置小波变换层能够提升网络训练效率与识别准确率,且小波变换第二层高频系数与原始图像结合输入卷积层时,网络模型效果最优。在不同光照条件下,相比于传统识别模型,该模型有更好的适应能力,对测试集的识别准确率达到93%。

The image feature values need to be manually selected and the robustness is poor in online coal-gangue separation.In order to improve the automation degree of coal-gangue separating, the original image coal and gangue collected on site was used as input to establish an image recognition model based on convolutional neural network. The convolutional neural network was visualized by deconvolution, and the process of extracting coal and gangue image features by convolutional neural network was analyzed. Through decomposing the original image by biorthogonal wavelet, the wavelet transform layer was set up in the convolutional neural network. The convolution operation was performed by combining the high frequency coefficient with the original image to optimize the recognition effect of the model. The results showed that this model effectively differentiated the coal and gangue images and had strong generalization ability.By setting the wavelet transform layer, the network training efficiency and recognition accuracy was improved.When combining the second layer high-frequency coefficient of wavelet transform with the original image, the network model was optimal. Compared with the traditional recognition model, the model had better adaptability under different illumination conditions, and the recognition accuracy of the test set reached 93%.

师亚文, 李务晋, 吕子奇. 煤矸图像识别网络的小波变换优化 [J]. 煤炭工程, 2023, 55(11): 160-166.