Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Object identification method

An object recognition and object technology, applied in the field of object recognition, can solve the problems of low accuracy, inability to identify arbitrary objects at any time, and limited types of object recognition.

Active Publication Date: 2021-11-16
广州微林软件有限公司
View PDF5 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

It has the characteristics of relatively few network model parameters, superior performance in speed, but slightly lower accuracy
In particular, the YOLO (You Look OnlyOnce) series has been updated to the fourth-generation algorithm YOLO v4 by 2020, which can achieve high recognition accuracy while maintaining the recognition speed, especially the lightweight network YOLO v4-tiny series, the number of model parameters is small, the speed is fast, and it is suitable for many industrial scenarios, but the accuracy still needs to be improved
[0008] Although the current target recognition algorithm can achieve high recognition and high speed, the types of object recognition are still limited. For example, YOLO9000 based on multi-level hierarchical structure can recognize 9000 kinds of objects, but it lacks flexibility and scalability and cannot be used at any time. to identify any item

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
  • Object identification method
  • Object identification method
  • Object identification method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0036] combine figure 1 As shown, this example provides a vision-based object recognition method and system, including three parts: object detection, matching weight matrix, and object recognition.

[0037] Preferably, the object detection method includes:

[0038] Step A1: Collect the target video stream through mechanisms such as sensors and infrared rays under different environmental scenarios, different weathers, and different lighting conditions, and use multimedia processing tools (such as FFmpeg) to process the target video frame to obtain multiple image frame sequences .

[0039] Step A2-1: Perform preprocessing on the acquired multi-image frame sequence. The preprocessing methods are not limited to filtering, screening, cropping, splicing, Gaussian noise and blurring. The preprocessed target object images constitute the target object dataset.

[0040] Step A2-2: Use Labelimg, a commonly used labeling tool for target detection, to label the target object to be detect...

Embodiment 2

[0093] Build a matching weight matrix.

[0094] Step B1, collecting related data sets.

[0095] Divide the target data set into two parts, one is the large class data set of the object, and the other is the small class data set of the object, and the classification standard can be divided according to attributes, shapes, categories, etc.

[0096]Step B2, performing relational processing on the target data set. The objects of the small category are associated with the objects of the large category. Each large category is associated with multiple subcategories. The relationship between the large category and the small category is a tree structure, such as Figure 6 As shown, a relationship hierarchical diagram is formed.

[0097] Step B3, according to the relationship hierarchical graph, establish a major category and minor category matching weight matrix through cosine similarity.

[0098] Specifically, the cosine similarity is,

[0099]

[0100] Among them, A and B are ...

Embodiment 3

[0102] Transform the MobileNetv2 recognition network to classify target objects.

[0103] The advantage of Mobilenetv2 is that it proposes Linear Bottleneck and Invered Residual.

[0104] Linear Bottleneck removes ReLU by removing the features of Eltwise+, reduces the damage of ReLU to features, and replaces the original nonlinear activation transformation with linear bottleneck (that is, does not use ReLU activation, but does linear transformation).

[0105] The above-mentioned Invered Residual changes the 3x3 convolution into a depth-separable convolution, which greatly reduces the amount of calculation, and can achieve more channel designs with better results. First increase the number of channels through 1 x 1 convolution, then use Depthwise's 3x3 spatial convolution and ReLU to alleviate the degradation of features by increasing the input dimension of ReLU, and finally use 1x1 convolution to reduce the dimension.

[0106] In order to pursue speed, Mobilenetv2 is deployed...

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 object identification method, which comprises the following steps: detecting an object, acquiring a related data set, calibrating an image of a target object data set by using an image marking tool for the acquired data set, performing data enhancement on the acquired data set, extracting a detected target object, and performing category classification, to classify the categories of target objects into a large category and a small category; and performing relation processing on the target data set to form a relation hierarchical graph, constructing large category and small category matching weight matrixes through the relation hierarchical graph and cosine similarity, and determining the category of an object and the object through an object identification network and the matching weight matrixes. The object identification method is accurate in identification, and can identify a plurality of objects of uncertain categories.

Description

technical field [0001] The invention relates to an object recognition method. Background technique [0002] Target detection and recognition is a basic problem in the field of computer vision. Fast and accurate positioning and recognition of specific targets in uncontrolled natural scenes is an important functional basis for many artificial intelligence application scenarios. [0003] In recent years, with the rapid development of deep learning technology, the target detection algorithm based on convolutional neural network has received attention and extensive research, and a large number of network structures with excellent performance and high efficiency have emerged, making the large-scale event application of the algorithm possible. . [0004] At present, the target detection algorithms emerging in academia and industry are divided into three categories. [0005] The first is the traditional object detection algorithm. For target confirmation based on sliding windows,...

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): G06K9/46G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/047G06N3/045G06F18/23213G06F18/2415G06F18/241
Inventor 张元本陈名国
Owner 广州微林软件有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com