Research On Environmental Sensing Technology Of Automotive Millimeter Wave Radar Based On Machine Learning

Autonomous driving has been recognized as one of the important directions for future automobile development.As a research hotspot of automotive system technology,autonomous driving technology has led to the development of a large number of related technologies such as vehicle positioning,interconnection,and detection.Among them,the automotive millimeter wave radar has become the core sensor supporting autonomous driving technology with its characteristics of full-time,all-weather,high integration,high performance and low cost.With the progress of artificial intelligence in recent years,the intelligent detection of automotive millimeter wave radar combined with artificial intelligence has provided strong support for L4 and even L5 autonomous driving.The perception and recognition of targets and environments is the main technical approach to promote accurate detection and optimization of control strategies.This paper aims at the key technical issues in auto-driving such as automotive millimeter-wave radar environment perception,scenes classification,and decisionmaking assistance,and carries out research on vehicle environment sensing and scenes classification technology based on machine learning.The main research contents are as follows:(1)The frequency-modulated continuous wave and MIMO detection mechanism widely used in automotive millimeter-wave radar are studied.The principle of MIMO angle measurement,the technology and process of frequency-modulated continuous wave detection are studied and analyzed,and the original method of raw data is introduced.(2)Research on scene classification and recognition technology based on support vector machine method.Utilizing the characteristics of different scene echo distributions,multiple statistics of radar one-dimensional range profile were extracted as classification features,and support vector machines were used to classify the scenes.On the premise of low computational load,it can achieve a classification accuracy rate of more than 80% on the test data set.(3)Based on the theory of neural networks,a deep neural network model based on a fully connected layer cascade and a convolutional neural network model based on a convolutional layer cascade were constructed,Both can reach 94% and 96% accuracy on the test set,respectively.Aiming at the problem of the effectiveness of the convolutional neural network for feature extraction,unsupervised dimensionality reduction processing was performed using the extracted features,which proved the ability of the neural network to extract information.Finally,the classification accuracy of the convolutional neural network algorithm on the test set can be increased to 98% by decision-level fusion.The above algorithms have been verified by measured data.The results show that the algorithm can classify the vehicle's environmental scenes effectively.The development of autonomous driving technology is inseparable from the participation of machine learning.Its self-learning and self-improving features can help in-vehicle sensors perceive known and unknown environments and provide reliable technical support for accurate decision-making of autonomous driving.