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

Multi-task cooperative scheduling method for online semantic segmentation machine vision detection

A technology of machine vision detection and semantic segmentation, which is applied in the direction of instruments, computer components, multi-program devices, etc., can solve problems such as long segmentation time, negative impact on online real-time performance, and large video memory usage, so as to improve real-time capabilities and reduce Storage overhead, the effect of improving computing resource efficiency

Pending Publication Date: 2021-10-15
SOUTH CHINA UNIV OF TECH
View PDF3 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0002] Machine vision systems based on deep learning semantic segmentation, in the case of high-resolution images, multi-network multi-images, detection and identification, etc., will have problems such as long segmentation time and large memory usage, which will negatively affect its online real-time performance

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
  • Multi-task cooperative scheduling method for online semantic segmentation machine vision detection
  • Multi-task cooperative scheduling method for online semantic segmentation machine vision detection
  • Multi-task cooperative scheduling method for online semantic segmentation machine vision detection

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0059] Implementation Example 1 of Cooperative Scheduling for Detection, Discrimination and Machine Vision Detection of Online Semantic Segmentation. Fully automatic detection of various interfaces, buttons and standard parts assembled on the front and rear panels of the ATX case. Specific requirements: ① Detect the entire front panel and rear panel of the chassis (width × height = 185 × 420mm), the positioning error of the interface and the button is ≤0.5mm, and the positioning error of the standard part is ≤0.2mm; The detection time of the chassis with more than 50 parts is ≤8s; ③For the assembly point with missing assembly or wrong assembly, its position and boundary information can be output.

[0060] The front panel and rear panel of the chassis have a large aspect ratio of 420 / 185≈2.27. The machine vision inspection system scheme is 3 industrial cameras with a pneumatic mechanism to collect images at different positions along the long sides of the front panel and rear pa...

Embodiment 2

[0069] Implementation example 2 of detection, discrimination and collaborative scheduling of online semantic segmentation machine vision detection. The development of bill anti-counterfeiting detector requires the realization of intelligent key anti-counterfeiting feature recognition of legal bills such as checks and bills of exchange. Specific requirements: ①Under 4 kinds of light excitation conditions including white light, backlight, infrared light, and ultraviolet light, detect 22 key anti-counterfeiting features such as watermarks, fluorescent main patterns, and emblems on complete bills; ②miniature characters on the amount column (1mm×1mm ), ticket number position anti-Stoke luminescence (line width about 0.1mm) and other micro-anti-counterfeit feature detection; Consistency of ticket number; ④ The detection time of single check and money order is ≤1s.

[0070] Considering that the scales of anti-counterfeiting features on the bills are quite different, one panoramic ca...

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 a multi-task cooperative scheduling method for online semantic segmentation machine vision detection. The method comprises the following steps: establishing a detection and identification cooperative scheduling task model; performing high-resolution image block semantic segmentation task scheduling: dividing each image into N sub-resolution Usub * Vsub, wherein the scheduling target is to obtain Usub, Vsub and Nsub values under the minimum value of the semantic segmentation time THigh of the high-resolution image; performing multi-network multi-image batch semantic segmentation task scheduling, packaging multiple images into Nbatch image groups, loading NCNN-GPU semantic segmentation network models on a GPU to carry out batch parallel processing, wherein the scheduling target is to obtain Nbatch and NCNN-GPU values under the minimum value of multi-network multi-image batch semantic segmentation time TLow; and scheduling the detection and identification cooperative scheduling task model to obtain an execution sequence under the minimum value of the total detection and identification time Tinspect, and completing detection and identification cooperative scheduling.

Description

technical field [0001] The present invention relates to the technical field of line machine vision detection based on deep learning, in particular to a multi-task cooperative scheduling method based on deep learning semantic segmentation. Background technique [0002] Machine vision systems based on deep learning semantic segmentation, in the case of high-resolution images, multi-network multi-images, detection and identification, etc., will have problems such as long segmentation time and large memory usage, which will negatively affect its online real-time performance. The multi-task parallel scheduling method is conducive to improving the segmentation time, video memory occupation and other indicators of online machine vision detection and identification. Among them, the underlying parallel scheduling method can optimize the computing efficiency of equipment, reduce the idle rate, and increase the parallel rate. It is necessary to select the appropriate underlying paralle...

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): G06F9/48G06T1/20G06K9/34G06K9/00G06N3/04G07D7/1205G07D7/128G07D7/20
CPCG06F9/4881G06T1/20G07D7/1205G07D7/128G07D7/2016G06N3/045
Inventor 刘桂雄黄坚
Owner SOUTH CHINA UNIV OF TECH