基于深度学习的矿山环境噪声识别方法OA
A method for identifying mine environmental noise based on deep learning
为了准确识别矿山环境噪声的种类,基于深度学习设计了一种噪声识别方法.首先,使用深度可分离卷积技术、残差连接与注意力机制设计了3种模块,通过组合这3种模块和传统的卷积神经网络模块构建了12种网络模型.然后,在Urbansound8k数据集的基础上增添了符合矿山环境的噪声种类,构建自制数据集,并在此数据集上对上述12种网络模型进行识别准确率测试.结果表明,所构建的12种模型的识别准确率均高于95%,效果最好的模型的识别准确率达到了96.8%.最后,利用基于该模型的噪声识别系统对全椒县马塘湖整合矿区的环境噪声进行识别,识别的平均准确率高于90%,有效实现了对矿山环境噪声种类的准确识别.
A noise identification method based on deep learning was designed to accurately identify the types of noise in the mining environment.First,three blocks were developed using depthwise separable convolution,residual connections and attention mechanisms.Next,12 network models were constructed by combining these blocks with traditional CNN block.Then,these 12 models were tested on a custom dataset which was created by adding mine-specific noise types to the Urbansound8k dataset.The results show that over 95%accuracy was achieved in all these 12 models and 96.8%accuracy was achieved in the best-performing model.Finally,the noise identification system based on this model was used to identify the environmental noise in the integrated mining area of Matanghu in Quanjiao County.The average accuracy rate of the identification was higher than 90%,effectively achieving the accurate identification of the types of environmental noise in the mining area.
韩龙成;刘福;廖同庆
安徽大学电子信息工程学院,安徽 合肥 230601安徽大学电子信息工程学院,安徽 合肥 230601安徽大学集成电路学院,安徽 合肥 230601
信息技术与安全科学
卷积神经网络噪声识别深度可分离卷积残差连接注意力机制
convolutional neural networknoise recognitiondepthwise separable convolutionresidual connectionattention mechanism
《安徽大学学报(自然科学版)》 2026 (5)
40-46,7
国家重点研发计划项目"零碳建筑(社区)评价关键技术研究与示范"(2022YFE0134100)
评论