高级检索
阮顺领, 景莹, 卢才武, 顾清华, 张雪飞. 基于深度卷积特征的露天矿卡车装载状况识别技术研究[J]. 煤炭科学技术, 2021, 49(10): 167-176.
引用本文: 阮顺领, 景莹, 卢才武, 顾清华, 张雪飞. 基于深度卷积特征的露天矿卡车装载状况识别技术研究[J]. 煤炭科学技术, 2021, 49(10): 167-176.
Shunling, JING Ying, LU Caiwu, GU Qinghua, ZHANG Xuefei. Study on recognition technology of truck loading condition in open-pitmine based on deep convolutional features[J]. COAL SCIENCE AND TECHNOLOGY, 2021, 49(10): 167-176.
Citation: Shunling, JING Ying, LU Caiwu, GU Qinghua, ZHANG Xuefei. Study on recognition technology of truck loading condition in open-pitmine based on deep convolutional features[J]. COAL SCIENCE AND TECHNOLOGY, 2021, 49(10): 167-176.

基于深度卷积特征的露天矿卡车装载状况识别技术研究

Study on recognition technology of truck loading condition in open-pitmine based on deep convolutional features

  • 摘要: 针对露天矿车辆运输过程中运载量管控受人为及环境等因素干扰较大,存在轻车跑票和人为套票等不利于生产管理的问题,提出了一种基于深度卷积特征的车辆装载状况识别方法。该方法通过构建试验数据集和对卷积神经网络AlexNet模型迁移学习,完成对露天矿卡车装载状况图像深度卷积特征的提取,并基于支持向量机多分类模型,实现对卡车装载状况的自动识别,在此基础上统计露天矿车队运载工作量。试验过程中,基于同一组试验数据集分别对GoogLeNet、ResNet、SqueezeNet、DenseNet模型进行迁移学习,提取卡车装载状况图像深度卷积特征,并使用同一支持向量机多分类模型对卡车装载状况进行自动识别。结果表明,在空间资源和时间资源约束下,迁移学习后的AlexNet模型在5种卷积神经网络中总体性能表现最佳,用其提取的图像深度卷积特征在卡车装载状态识别中准确率最高。相比于传统的人工设计图像特征,该方法能够更好地完成露天矿卡车装载状况自动识别任务,试验数据集的识别准确率达到97%以上,在此基础上对露天矿车队运载工作量进行统计,可有效鉴别露天矿卡车的实际装载状况,提高露天矿卡车运载的吨公里生产效率,有效解决露天矿山车辆运载工作量的管控问题。

     

    Abstract: In view of the large interference ofman-made and environmental factors in the transportation of open-pit mine vehicles, the control of the load is greatly affected by human and environmental factors, and there are problems such as light vehicle ticketing and man-made packages that are not conducive to production management, a method for open-pit mine truck loading condition recognition was proposed based on deep convolution features.By construction of experimental dataset,the deep convolutional features can be extracted from open-pit mine truck loading condition images by using transfer learning AlexNet model,and the automatic recognition of truck loading condition can be realized based on SVM(Support Vector Machine) multi-classification model.Then,open-pit mine fleet workload statistics can be completed on this basis.During the experiment, based on the same set of experimental data sets, the GoogLeNet, ResNet, SqueezeNet, and DenseNet models were subjected to migration learning, and the deep convolution features of the truck loading condition image were extracted, and the same support vector machine multi-classification model was used to automatically recognize the truck loading condition .The results show that under the constraints of space and time resources, the AlexNet model after transfer learning has the best overall performance among the five convolutional neural networks, and the image deep convolution features extracted by it have the highest accuracy in truck loading state recognition. Compared with the traditional manual design of image features, this methodcan better complete the automatic identification task of truck loading status in open-pit mines, and the recognition accuracy of the test data set is more than 97%. On this basis, statistics on the carrying workload of the open-pit mine fleet can effectively identify the actual loading conditions of the open-pit trucks, improve the production efficiency of the ton-kilometers carried by the open-pit trucks, and effectively solve the problem of management and control of the open-pit mine vehicle workload.

     

/

返回文章
返回