Unlock instant, AI-driven research and patent intelligence for your innovation.

Character detection method and system based on transfer learning

A technology of transfer learning and detection methods, applied in the direction of integrated learning, biometric recognition, character and pattern recognition, etc., can solve problems such as students' network misleading, and achieve the effect of improving accuracy

Pending Publication Date: 2022-01-14
INST OF COMPUTING TECH CHINESE ACAD OF SCI
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Mimicking features that are mistaken for context by the teacher network can be misleading for the student network

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
  • Character detection method and system based on transfer learning
  • Character detection method and system based on transfer learning
  • Character detection method and system based on transfer learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0041] To solve the above problems, the present invention proposes a Feature Richness Score (FRS) method to select important features that are beneficial for distillation. Feature richness refers to the amount of information about objects contained in features, and can be expressed by the probability that these features are objects. Extracting features with high feature richness instead of features in the bounding box area can effectively solve the above two limitations - ignoring the features of objects outside the bounding box that are not included in the category of the dataset; and paying too much attention to the misclassification of the teacher detector Characteristics.

[0042] First, the features of objects not included in the categories of the dataset have high feature richness. Therefore, important features outside the bounding box can be retrieved using feature richness, which can guide the student network to learn the generalized detectability of the teacher netwo...

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 character detection method and system based on transfer learning, and the method comprises the steps: constructing a teacher network for image target detection and a student network corresponding to the teacher network, and obtaining a picture data set marked with character category labels as a training data set; when the training data set is used to train the teacher network and the student network, extracting the classification branch output of each layer of target detection FPN of the teacher network, and obtaining a four-dimensional matrix including NCHW four channel output results; summing in the C channel direction of the four-dimensional matrix to obtain a feature matrix of an NHW dimension as a feature mask matrix, and obtaining feature map constraint loss based on the feature mask matrix and FPN feature maps of the teacher network and the student network; and summing the loss of the teacher network, the loss of the student network and the constraint loss of the feature map to obtain distillation loss, and after the distillation loss converges, using the student network to detect the figure in the picture.

Description

technical field [0001] The present invention relates to the technical field of knowledge distillation in target detection and transfer learning, and in particular to a transfer learning-based person detection method, system, storage medium and client. Background technique [0002] In recent years, large-scale deep models have achieved great success, but the enormous computational complexity and massive storage requirements make their deployment in resource-constrained devices a great challenge. As a model compression and acceleration method, knowledge distillation effectively improves the performance of small models by transferring the dark knowledge from the teacher detector, that is, information implicit in the teacher network that is useful in the student network. Most existing object detection methods based on knowledge distillation mainly let the student network imitate the features overlapping with the bounding box in the teacher network, and consider the foreground fe...

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): G06V40/10G06V10/764G06V10/774G06V10/70G06K9/62G06N20/20
CPCG06N20/20G06F18/214G06F18/241
Inventor 张蕊杜治兴常明张曦珊刘少礼
Owner INST OF COMPUTING TECH CHINESE ACAD OF SCI