Construction method of multi-task classification network based on orthogonal loss function

A loss function and classification network technology, applied in neural learning methods, biological neural network models, neural architectures, etc., can solve problems such as lack of common features and unique feature distinctions, and achieve the effect of improving separability and suppressing interference

Active Publication Date: 2020-03-27
NORTHWESTERN POLYTECHNICAL UNIV
View PDF3 Cites 8 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, due to the sharing of hidden layer parameters, the extracted features are intersected. Therefore, the multi-task classification netwo

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
  • Construction method of multi-task classification network based on orthogonal loss function
  • Construction method of multi-task classification network based on orthogonal loss function
  • Construction method of multi-task classification network based on orthogonal loss function

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0032] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0033] Step 1: Build a hierarchical label tree;

[0034]Because there are visual similarities between species belonging to the same race, a hierarchical label tree is constructed using the affiliation relationship between species in nature for different databases. The structure of the hierarchical label tree is divided into two layers. The first layer of labels is the coarse category label of the image, which is divided according to the species the image belongs to. The second layer label is the fine category label of the image, which is defined according to the subcategory of which species the image belongs to. The branches of the tag tree represent affiliation relationships. Caltech-UCSD Birds-200-2011 is a bird database, which is constructed by consulting encyclopedia knowledge and dividing rough categories according to the families and genera of bi...

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 construction method of a multi-task classification network based on an orthogonal loss function. The constructed multi-task classification network simulates a human learning process, a deep convolutional neural network is used as a hidden layer to simulate the brain of a human to perform deep feature extraction, a tree classifier is used as a task-related output layer to perform progressive classification, and the recognition process forms different learning tasks Characteristics obtained by different tasks to better meet respective requirements. When the classifier completes a coarse classification task, the depth features of the same coarse class are more aggregated; and when the fine classification task is completed, the depth features of different fine classesare more discrete, and the task output layer features of different classification tasks are distinguished, so that classifiers of different hierarchies can obtain features better matched with different classification tasks, useless features are removed, and the classification accuracy is improved.

Description

technical field [0001] The invention relates to the field of image classification, in particular to a multi-task classification construction method. Background technique [0002] In recent years, photo recognition has become more and more widely used in field exploration and daily life. This is thanks to the development of deep learning. The current best-performing tool for feature extraction is the deep convolutional neural network. As we all know, the deep convolutional neural network can not only extract edge information in the shallow layer, but also the semantic information of the feature will become more and more abstract as the number of layers deepens, and the obtained features are closer to human cognitive behavior. [0003] Subsequently, multi-task classification networks have gradually entered people's sight. The different tasks of the multi-task classification network assist each other, and the different tasks of the same network are trained at the same time, ...

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): G06K9/00G06K9/62G06N3/04G06N3/08G06N20/00
CPCG06N3/084G06N20/00G06V20/10G06N3/045G06F18/241G06F18/24323
Inventor 何贵青敖振霍胤丞纪佳琪
Owner NORTHWESTERN POLYTECHNICAL 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