Research on unloading drill-rod action identification in coal mine water exploratio
-
摘要: 针对煤矿井下探水作业监工人员通过观看视频来监控卸杆作业的方式存在效率低下且极易出错的问题,提出利用三维卷积神经网络(3DCNN)模型对探水作业中的卸杆动作进行识别。3DCNN模型使用3D卷积层自动完成动作特征提取,通过3D池化层对运动特征进行降维,通过Softmax分类处理来识别卸杆动作,并使用批量归一化层提高模型的收敛速度和识别准确率。采用3DCNN模型对卸杆动作进行识别时,首先对数据集进行预处理,从每段视频中均匀抽取几帧图像作为某动作的代表,并降低分辨率;然后采用训练集对3DCNN模型进行训练,并保存训练好的权重文件;最后采用训练好的3DCNN模型对测试集进行测试,得出分类结果。实验结果表明,设置采样帧数为10帧、分辨率为32×32、学习率为0.000 1,3DCNN模型对卸杆动作的识别准确率最高可达98.86%。Abstract: In view of low efficiency and error prone problems in the way that supervisors of underground water exploration operation realize monitoring of unloading drill-rod operation by watching video, 3D convolutional neural network (3DCNN) model is proposed to identify unloading drill-rod action in water exploration operation. In 3DCNN model, 3D convolution layer is used to automatically extract action features, 3D pooling layer is used to reduce dimension of motion features, softmax classification is used to identify unloading dirll-rod action, and batch normalization layer is used to improve convergence speed and identification accuracy of the model. When the 3DCNN model is used to identify unloading drill-rod action, firstly, the data set is preprocessed, and several frames of images are extracted from each video as representatives of an action, and the resolution is reduced; secondly, the training set is used to train the 3DCNN model, and the trained weight file is saved; finally, the trained 3DCNN model is used to test the test set, and the classification results are obtained. The experimental results show that when the number of sampling frames is 10, the resolution is 32×32, and the learning rate is 0.000 1, the highest recognition accuracy of the model can reach 98.86%.
点击查看大图
计量
- 文章访问数: 67
- HTML全文浏览量: 12
- PDF下载量: 14
- 被引次数: 0