Machine vision recognition system based on artificial error correction mechanism and deep learning network collaboration

Abstract

Based on the artificial error correction mechanism and the deep learning network collaborative machine vision recognition system, the deep learning network receives the captured visual feedback images in real time, performs deep learning processing on each frame of the received image, and converts the processed image and the coordinates of the detection target on the image, The angle and time information is sent to the intermediate result comprehensive processing unit; the processed image is sent to the man-machine error correction interface; the man-machine error correction interface continuously displays the processed image of each frame received, and the visual feedback image is manually captured According to the specific situation, remove the emphatic display of the obvious wrong target on the displayed image, and send the relevant information to the intermediate result comprehensive processing unit; the intermediate result comprehensive processing unit detects the target from the received frame of image The information of the wrong target is eliminated from the coordinates, angle and time information of the image coordinate system, and the coordinates, angle and time information of the recognized target are obtained. At the same time, the emphasis display of the wrong target is removed from the received visual feedback image, and the completion Visual Identity.

Description

technical field [0001] The invention relates to a machine vision recognition system based on artificial error correction mechanism and deep learning network cooperation. Background technique [0002] In machine vision technology, the existing image recognition methods usually extract certain features, compare the obtained features with ideal values, and compare the high similarity as the recognition result. [0003] In the field of multi-target sorting, there are many target categories and feature types that need to be identified, such as bottles and cans of various shapes and colors. Due to the large amount of calculation of feature extraction, the method of extracting multiple features cannot be achieved in real time. Meet the requirements; and the accuracy of the feature extraction method has not been high, and it is difficult to meet the use requirements in the industrial automation production line. Contents of the invention [0004] The technical solution of the pres...

AI Technical Summary

Problems solved by technology

[0003] In the field of multi-target sorting, there are many target categories and feature types that need to be identified, such as bottles and cans of various shapes and colors. Due to the large amount of calculation of feature extraction, the method of extracting multiple features cannot be achieved in real time. Meet the requirements; and the accuracy of the feature extraction method has not been high, and it is difficult to meet the use requirements in the industrial automation production line

Images