A Vehicle Identification Method Based on Deep Network Model of Spatial Pyramid Pooling

Abstract

The invention discloses a car model recognition method based on a deep network model of spatial pyramid pooling. First, an image of a car model database is imported into a deep network model to extract features of a convolutional layer to form a feature map of the convolutional layer; For each image of the feature map, the spatial pyramid convolution operation is performed according to different scales to form the feature map of the spatial pyramid layer; then all the features of the spatial pyramid layer are pooled to form a fully connected layer to obtain the final feature representation of the car image; the car model The feature representation of the image is used for linear support vector machine training to obtain a vehicle type recognition system; the vehicle to be recognized is also obtained from its feature representation, and then imported into the recognition system to identify its vehicle type. The input image of the traditional deep network model must be a fixed size, which limits the operation of large-scale vehicle image data. The present invention adopts a deep network model based on spatial pyramid pooling, which solves this problem well and has high practicability and robustness.

Description

technical field [0001] The invention belongs to the technical field of machine learning, pattern classification and recognition, and in particular relates to a car model recognition method based on a deep network model of spatial pyramid pooling. Background of the invention [0002] With the continuous improvement of living standards in modern society and the rapid growth of the number of cars, traffic supervision is facing great challenges. As an important means of traffic supervision, video surveillance system has been widely used in various fields of modern traffic. However, the traditional method of relying on manual interpretation can no longer meet the needs of today's massive traffic video processing, and it has become an inevitable trend to build an intelligent recognition system to automatically process various traffic video information. The identification of vehicle types in traffic video images, as a key technology in construction, has long been widely concerned ...

AI Technical Summary

Problems solved by technology

[0004] However, the size of the input image of the traditional convolutional neural network architecture is fixed (for example: 256x256), and this artificially changing the size of the input image destroys the scale and aspect ratio of the input image, so there are problems: (1 ) The choice of scale is subjective. For different targets, the most suitable size may be different; (2) For images of different sizes and aspect ratios, forcing the transformation to a fixed size will lose information; (3) The forced transformed image may not contain a complete image, and may also cause geometric deformation of the image

Images