高速列车自主视觉定位的测试方法研究

Research on Test Method for Autonomous Visual Positioning of High-speed Trains

列车定位是保障高速列车安全运行的关键技术。随着“故障-安全运行”需求的提出,基于深度学习的列车自主视觉定位技术受到了广泛关注。本文针对列车定位的安全性问题,提出了一种神经网络模型测试方法,通过图像变换得到测试案例,并通过深度变异方法对测试案例进行量化评价。实验结果表明,该方法产生的测试案例能够有效暴露出不同场景下神经网络模型的识别错误,为提升新型列.

Train location is the citical technology to ensure the safe operation of high-speed trains.With the de-mand for"fail-safe operation",the autonomous visual positioning technology of trains based on deep learning hasattracted wide attention.Aiming at the safety problem of train location,this paper proposed a neural networkmodel testing method.This approach could generate test cases by image transformation,and evaluate them by using the deep variation method.Experimental results show that the test cases generated by this method can effec-tively expose the identification erors of neural network models in different scenarios,and provide a basis for im-proving the safety of the new train autonomous visual positioning technology.