Convolutional neural network training method based on network image

A convolutional neural network, network image technology, applied in neural learning methods, biological neural network models, neural architectures, etc., can solve the problems of limited number of images, accuracy can not reach the standard, and recognition model recognition categories are insufficient.

Pending Publication Date: 2021-07-06
GOSUNCN TECH GRP
View PDF9 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0017] Disadvantages of prior art: These methods are all based on a database with limited image elements, which means limited categories for convenience and limited number of images on the other hand
The neural network needs a large amount of data for training to get the best results, and the limitation of this database leads to insufficient recognition categories of the trained recognition model, and the accuracy cannot meet the standard.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Convolutional neural network training method based on network image
  • Convolutional neural network training method based on network image
  • Convolutional neural network training method based on network image

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0054] Example 1: Web image crawling

[0055] It is difficult to label the brand of the vehicle captured in the surveillance image, and at the same time, there are vehicle images with good brand label information on the Internet. It is hoped that the brand-labeled image of the network will be used as the initial data for the training of the vehicle brand recognition model. Therefore, it is first necessary to crawl the high-quality labeled vehicle images on the Internet. Crawl the image data of vehicles marked with vehicle brands by crawling pictures on autolist.com, autohome.com and other websites such as automobile information and second-hand car sales networks.

Embodiment 2

[0056] Example 2: Iterative data screening and model training

[0057] Due to the great difference between the network image and the surveillance image (the feature distribution is inconsistent), it is not good to directly use the model trained by the network image to recognize the surveillance image, so a gradual data screening and model training is adopted. method. For the initial stage, network images are used as the training set to train a vehicle brand recognition model, which is then applied to surveillance images. For each input image, the recognition model outputs a class probability distribution, where a value in the distribution represents the probability that the image contains a number of vehicles belonging to a certain class. The image is filtered according to a preset threshold. Only keep images with the largest value in the distribution greater than a preset threshold. For these retained images, the category represented by the maximum value is also added to t...

Embodiment 3

[0058] Embodiment 3: basic feature extraction

[0059] For an image that contains the back shot of the vehicle, the size of the image is adjusted to 256*256 by linear interpolation, and then input into the convolutional neural network, and the output of the model is aggregated through a layer of convolutional network layers. The output of this layer is the extracted image representation. The representation is a one-dimensional vector with 2048 elements.

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention provides a convolutional neural network training method based on network images. A network car purchase website has a large number of rear-shot images of cars with category labels, the images can provide basic car brand information, but feature distribution of the images is different from that of car images shot by real monitoring, so that the images cannot be well applied to real monitoring scenes. Therefore, the recognition model is trained in advance through the network image and then applied to the monitoring image, and the image with the high score in the monitoring image is used as the training data to train the recognition model again, so that the effect of continuously promoting the recognition accuracy is achieved.

Description

technical field [0001] The invention belongs to the technical field of image recognition, and in particular relates to a convolutional neural network training method based on network images. Background technique [0002] Vehicles are one of the most important means of transportation and logistics in modern times. They occupy a considerable proportion in urban life and are closely related to people's lives. However, with the development of modern automobile production technology, the number and brands of vehicles are gradually increasing, so intelligent vehicles are needed. means of management. The construction of the urban monitoring system has gradually increased the number of cameras in the city, which can capture vehicle image information very well. The post-shot images of vehicles tend to have higher similarity, and the differences between different categories are smaller. Different types of vehicles only have subtle differences in the position of lights, etc. There ar...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/00G06K9/62G06N3/04G06N3/08G06T3/40
CPCG06N3/084G06T3/4023G06V20/54G06V2201/08G06N3/045G06F18/214
Inventor 林焕凯傅慧源陈利军洪曙光马东华王川铭董常青王祥雪刘双广
Owner GOSUNCN TECH GRP
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap