Target labeling method and target labeling device

Abstract

The invention discloses a target labeling method and a target labeling device. The method comprises the steps of: using a trained target detection model for labeling a prediction picture set, and obtaining prediction box information, wherein the prediction picture set comprises labeled pictures and unlabeled pictures; from the prediction frames, screening out the prediction frames of which the coincidence degree with the labeled target is lower than a preset value or which are lack of the labeled target; classifying the screened prediction frames by using a built and trained classification network, wherein the classification network is obtained by training the marked target and a background image thereof; and if the classification result of the prediction box is consistent with the prediction category of the prediction box, writing the prediction box information into a labeling file. According to the invention, the trained model is used for labeling the prediction picture set, the prediction frame of which the coincidence degree with the labeled target is lower than the preset value or lacks the labeled target is screened out, and the classification network is used for classification inspection, so that intelligent implementation and result inspection of target labeling are realized, and the efficiency and the accuracy of target labeling can be improved.

Description

technical field [0001] The present invention relates to the technical field of object labeling, in particular to an object labeling method and an object labeling device. Background technique [0002] In recent years, traditional manufacturing is gradually transforming into intelligent manufacturing. Traditional manual labeling tends to reduce accuracy and efficiency as the fatigue of labelers increases, and manual labeling tends to be unstable with changes in judgment standards. First of all, the standard of human judgment cannot guarantee continuous stability. Secondly, human beings are easily affected by the heavy workload, resulting in fatigue and a decline in enthusiasm, which leads to a decrease in the accuracy of the work. Need to mark) In front of the work that needs to be marked, it needs to be realized with the help of intelligent methods. Deep learning is an important part of intelligent manufacturing, so the correct and fast labeling of images used for deep lear...

AI Technical Summary

Problems solved by technology

Traditional manual labeling tends to reduce accuracy and efficiency as the fatigue of labelers increases, and manual labeling tends to be unstable with changes in judgment standards

First of all, the standard of human judgment cannot guarantee continuous stability. Secondly, human beings are easily affected by the heavy workload, resulting in fatigue and a decline in enthusiasm, which leads to a decrease in the accuracy of the work. need to be marked) before the work that needs to be marked, it is necessary to use intelligent methods to realize

Images