An image classification network training method based on massive single-class and single-amplitude images

A single image and classification network technology, applied in pattern recognition, can solve problems such as unsatisfactory effect, loss of effect of DeepID, no intra-class distance, etc., to increase inter-class distance, reduce intra-class distance, and recognition accuracy high effect

Active Publication Date: 2019-02-12
CHINA JILIANG UNIV
View PDF5 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, DeepID uses the face ID card dataset to conduct experiments, and the effect is not very satisfactory.
This is because the face

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
  • An image classification network training method based on massive single-class and single-amplitude images
  • An image classification network training method based on massive single-class and single-amplitude images
  • An image classification network training method based on massive single-class and single-amplitude images

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0024] In this embodiment, the process is as follows figure 1 As shown, a flow chart of image classification network training based on massive single-category single-frame images. The specific implementation mainly includes the following steps:

[0025] Step (1): Data preparation: Use 1 million single-type and single-face ID card datasets as training dataset 1, and use Webface face dataset as training dataset 2.

[0026] Step (2): Data preprocessing: convert the prepared images of the two face datasets into 128*128 resolution.

[0027] Step (3): Alternate cycle training is based on the image classification network of massive single class single image: Mini-batch is the number of data samples processed in each batch of training, and the present invention adopts that the number of training in each batch is 64, when When the number of training iterations is odd, randomly select 64 pictures from dataset 1 without repetition, and input them into the image classification network. I...

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 discloses an image classification network training method based on massive single-class and single-amplitude images. The training data sets of double data forms of the single-class and single images and the single-class and multi-amplitude images are used to train an image classification network of the massive single-class and single-amplitude images alternately and cyclically, a training data input layer is replaced with two network layers of one input layer of a training data set 1 and two input layers of a training data set 2. When the number of training iterations is odd, thetraining data set is used as the input data of the image classification network base on the massive single-class and single-amplitude images, a dynamic loss function based on iteration times adopts an inter-class distance loss function to train the network. When the number of iterations is even, the training data set 2 is used as the input data of the image classification network base on the massive single-class and single-amplitude images, and the dynamic loss function based on the iteration times combines the center loss and Soft-max loss functions as the loss functions of the training network to train the network, thereby obtaining an image classification model.

Description

technical field [0001] The invention relates to technical fields such as computer vision, pattern recognition, and machine learning, in particular to an image classification network training method based on a single-type single-frame image. Background technique [0002] Machine learning methods are widely used in image analysis to complete specific tasks on new data, such as classification, recognition, and segmentation, by training a model on a given data set. Commonly used algorithms include Support Vector Machine (SVM), Hidden Markov (HMM) and Artificial Neural Network. However, traditional machine learning algorithms rely on prior knowledge to manually extract features from raw data to train models. Due to the difficulty of feature selection, the model may have overfitting problems, and the generalization ability is difficult to guarantee; on the other hand, traditional models are difficult to adapt to large-scale data sets, and the model scalability is poor. [0003] ...

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
IPC IPC(8): G06K9/62G06K9/46
CPCG06V10/40G06F18/241G06F18/214
Inventor 章东平郑寅陶禹诺陈思瑶毕崇圆
Owner CHINA JILIANG UNIV
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