Vehicle type recognition method based on CNN multiple-features combining and multi-core sparse representation

Abstract

A vehicle type recognition method based on CNN multiple-features combining and multi-core sparse representation includes the steps of using a smart traffic camera to take a vehicle image under complex scene, conducting pretreatment for the vehicle image, designing five convolutional layers and three full connection layers and automatically extracting global and local characteristics of the vehicle based on AlexNet network in CNN, taking the feature maps of the first, second and fifth pooling layers and sixth and seventh full connection layers subjected to column stretching as vehicle feature components and inputting to five different single-core functions, forming a combined feature matrix through weighted fusion, solving the weight of each function and the projection matrix of the combined feature matrix, and realizing vehicle type recognition according to core sparse minimal reconstruction error. Real-time vehicle type identification can be realized, traffic flow information needed can be extracted, traffic jam problem can be solved, and general planning and road construction of a road network are facilitated.

Description

technical field [0001] The invention belongs to the field of vehicle vehicle identification in intelligent transportation systems, and in particular relates to a vehicle vehicle identification method based on CNN multi-feature combination and multi-core sparse representation. Background technique [0002] Convolutional neural network (CNN) has become a research hotspot in the field of image recognition. Due to its high recognition rate and other advantages, convolutional neural network has gradually been applied to vehicle recognition. But in general, the deep extracted features of the convolutional neural network are used for classifier training, and these features may not contain enough useful information to achieve the correct classification of the image. Studies have shown that the shallow layer of the convolutional neural network extracts the local features of the image, which are finer and contain more detailed information, while the deep layer extracts the global feat...

AI Technical Summary

Problems solved by technology

[0003] Kernel learning methods have been successfully applied in the field of image processing, but most kernel learning methods are based on single-core methods, because the performance of different kernel functions varies greatly, and single-kernel learning methods cannot meet the needs of data heterogeneity, data inconsistency, etc. Application requirements under conditions such as rules and uneven samples cannot effectively integrate the hidden information behind heterogeneous data

Images