Target detection method and device based on semi-supervised learning and storage medium

Abstract

The invention relates to the technical field of target detection, and discloses a target detection method based on semi-supervised learning, and the method comprises the steps: determining label datacorresponding to training data based on the obtained training data; performing data cleaning processing on the label data to obtain cleaned new label data; performing data enhancement processing on the new label data to obtain enhanced data corresponding to the new label data; training a deep learning model based on the enhanced data and preset manually annotated image information until a loss function of the deep learning model converges within a preset range to form a target detection model; and obtaining a target detection result of the to-be-detected data based on the target detection model. The invention also relates to a blockchain technology, and the new label data is stored in the blockchain. According to the invention, the target detection efficiency and accuracy based on semi-supervised learning can be improved.

Description

technical field [0001] The present invention relates to the technical field of target detection, in particular to a method, device, electronic equipment and computer-readable storage medium for target detection based on semi-supervised learning. Background technique [0002] The artificiality behind artificial intelligence mainly refers to the need for a large amount of manpower to label data before training the model. Although there are currently public target detection datasets such as COCO, before the target detection depth model is applied to actual projects, it still needs to be retrained on the labeled business dataset to adapt to the business data. At present, most artificial intelligence enterprises need to invest a lot of cost in obtaining manual labeling of business data. At the same time, for the data that has been marked, it is also necessary to invest in manual inspection, cleaning, and correction to ensure the quality of image marking. This demand comes from t...

AI Technical Summary

Problems solved by technology

At present, most artificial intelligence companies need to invest a lot of cost in obtaining manual annotation of business data

At the same time, for the data that has been marked, it is also necessary to invest in manual inspection, cleaning, and correction to ensure the quality of image marking. This demand comes from the sensitivity of neural networks to data, so labeling data needs to build a multi-level labeling and review structure , for large batches of data, it is often only through sampling inspection to prove that statistics are available

[0003] At present, although the semi-supervised learning method for classification tasks has achieved certain results, the semi-supervised learning method for target detection is not yet mature, and there are still problems such as calculation accuracy and large amount of data.

Images