基于AI语义分割的内燃机车自动驾驶在煤炭装车中的研究

Research on automatic driving of diesel locomotive in coal loading based on AI semantic segmentation

煤炭定量装车站智能化装车需要机车慢速且恒速运行,目前煤炭装车系统完全依靠人工通过对讲机以喊话的方式来告知机车驾驶员控制车速。随着人工智能技术的发展,在装煤机车控制中,可以实现自动加速。其中,自动驾驶首要任务就是判断前方可行驶区域。受UNet和Segnet网络启发,设计了一种基于AI语义分割技术的铁轨分割模型。首先采集列车视角的铁轨数据,对数据预处理和标注,制作装车站环境下的铁轨数据集。然后构建Encode-Decode结构的神经网络,Encode阶段以卷积残差网络为主干网络,提取并融合不同层级的特征图。Decode阶段结合特征图和最大池化索引通过转置卷积恢复图像细节特征。最后通过两阶段随机梯度下降优化算法训练网络,得到铁轨分割模型。网络模型训练完毕以后,在数据集上进行对比试验,实验结果表明,该方法相对传统图像检测的方法可以有效分割出铁轨区域,模型泛化能力较强。

The intelligent loading of coal loading station requires the locomotive to run at a slow and constant speed.At present,the coal loading system relies on manual interphone to tell the locomotive driver to control the speed.With the development of artificial intelligence technology,automatic acceleration can be realized in the control of coal loading locomotive.Among them,the first task of automatic driving is to judge the driving area ahead.Inspired by unet and segnet,this paper designs a rail segmentation model based on AI semantic segmentation technology.First,collect the rail data from the train perspective,preprocess and label the data,and make the rail data set under the loading station environment.Then,an encoding code structure neural network is constructed.In the Encoding stage,convolutional residual network is used as the backbone network to extract and fuse the feature maps of different levels.Decode stage combines feature map and maximum pool index to recover image details through transpose convolution.Finally,the network is trained by two-stage random gradient descent optimization algorithm,and the rail segmentation model is obtained.After the training of the network model is completed,a comparative experiment is carried out on the data set.The experimental results show that this method can effectively segment the rail area compared with the traditional image detection methods,and the model generalization ability is strong.