A method for road surface type estimation based on deep convolutional neural network without loss function

Abstract

The invention discloses a road surface type estimation method based on a deep convolutional neural network without a loss function, comprising step 1, collecting road surface working condition images, calibrating the road surface type and establishing a road surface working condition database; the image is based on a lossless function The deep convolutional neural network is used for training, image features are obtained, and binary hash coding and histogram processing are performed to obtain the feature output vector of the image; according to the feature output vector of the image and its corresponding road type, the support vector machine is trained and Select parameters to determine the road surface type discriminant function; Step 2, collect the image of the road surface to be tested, and obtain the characteristic output vector of the road surface to be tested according to the step 1, and use the trained support vector machine to determine the type of road surface to be tested. It simplifies the feature extraction of the image by the deep learning model of the convolutional neural network, and uses the support vector machine to classify the image, which greatly reduces the difficulty of convolution training and improves the classification efficiency.

Description

technical field [0001] The present invention relates to the field of automobile powertrain control, and more specifically, the present invention relates to a road surface type estimation method based on a deep convolutional neural network without loss function. Background technique [0002] The road surface adhesion coefficient is an important parameter of the vehicle's active safety control strategy. Assuming that the parameter value can be estimated in real time, the control system can adjust the control strategy in real time according to the current road surface conditions and vehicle driving status, so as to avoid traffic accidents caused by poor road adhesion conditions. Accidents, improve the safety, handling, economy and comfort of the car. [0003] At present, most scholars at home and abroad use commonly used on-board sensors to measure the motion response of the car body or wheels when driving on different roads to estimate the road adhesion coefficient, but this e...

AI Technical Summary

Problems solved by technology

[0003] At present, most scholars at home and abroad use commonly used on-board sensors to measure the motion response of the car body or wheels when driving on different roads to estimate the road adhesion coefficient, but this estimation method is to estimate the road adhesion coefficient after the tire touches the road surface, and All kinds of methods are either capable of identifying a single type of road surface or are limited by estimation conditions or estimation conditions, etc.

Images