Scholarlink

<p>As one of the important components in the roof support operation of roadways, the pallet can achieve automatic positioning of equipment, monitoring of roadway deformation and abnormal monitoring of anchor rods in recent years by identifying the pallet. Among them, whether pallets can be identified accurately and in real time is the key to the reliability of this type of application. Faster R- CNN network is used as the architecture to optimize the network in terms of both recognition accuracy and recognition speed. The recognition accuracy of pallets is enhanced by selecting ResNet50 as the backbone network, optimizing the size of candidate boxes in the regional suggestion network, and integrating an attention mechanism module into the feature extraction network. Introduce depthwise separable convolution and knowledge distillation algorithms to maintain the accuracy of pallet recognition while enhancing the speed of pallet recognition. The test results indicate that the mAP value of the improved model proposed in this paper reaches 92.01%, which is 7.07% higher than that of the original model. The processing time of each image is 15.6 ms, demonstrating the potential of the proposed method for real-time pallet recognition under the simulated conditions considered in this study. The proposed method provides technical support for future intelligent applications and further validation in underground coal-mine scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on roadway pallet recognition method for complex scenarios in underground coal mines

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Lisen Ma,
  • Yuan Tian,
  • Mingrui Hao,
  • Lei Zhang,
  • Rui Qiao,
  • Yueqi Bi,
  • Qu Jia,
  • Xiaoyan Li,
  • Jiaran Li

摘要

As one of the important components in the roof support operation of roadways, the pallet can achieve automatic positioning of equipment, monitoring of roadway deformation and abnormal monitoring of anchor rods in recent years by identifying the pallet. Among them, whether pallets can be identified accurately and in real time is the key to the reliability of this type of application. Faster R- CNN network is used as the architecture to optimize the network in terms of both recognition accuracy and recognition speed. The recognition accuracy of pallets is enhanced by selecting ResNet50 as the backbone network, optimizing the size of candidate boxes in the regional suggestion network, and integrating an attention mechanism module into the feature extraction network. Introduce depthwise separable convolution and knowledge distillation algorithms to maintain the accuracy of pallet recognition while enhancing the speed of pallet recognition. The test results indicate that the mAP value of the improved model proposed in this paper reaches 92.01%, which is 7.07% higher than that of the original model. The processing time of each image is 15.6 ms, demonstrating the potential of the proposed method for real-time pallet recognition under the simulated conditions considered in this study. The proposed method provides technical support for future intelligent applications and further validation in underground coal-mine scenarios.