A weakly interactive object detection deep learning method and system

Abstract

The present invention discloses a weakly interactive deep learning method and system for object detection. The method includes: step S1, selecting some unlabeled data for manual labeling of object recognition, and combining it with the public data set to form an initial data set; step S2, select a deep learning model, and use the labeled data in step S1 to train the model; step S3, use the model obtained from training to perform feature extraction on the unlabeled data and labeled data of the initial data set; step S4, the Combining the features, establishing a feature matrix, using the labeled data to perform label mapping on the unlabeled data, and mapping the label to the unlabeled data, and completing the labeling of the unlabeled data; step S5, according to the results of step S4 and step S1 The labeled data are combined into a new labeled data training set; step S6, using the new labeled data training set to repeatedly train the model until the model performance reaches the expected effect.

Description

technical field [0001] The invention relates to the technical field of deep neural networks, in particular to a weak interactive deep learning method for object detection and a system thereof. Background technique [0002] Image object classification and detection are two important basic problems in computer vision research. They are also the basis for other high-level vision tasks such as image segmentation, object tracking, and behavior analysis. They are very active research directions in the fields of computer vision, pattern recognition, and machine learning. . Object classification and detection are widely used in many fields, including face recognition, pedestrian detection, intelligent video analysis, pedestrian tracking, etc. Content-based image retrieval in the Internet field, automatic classification of albums, etc. The computer automatic classification and detection technology has also reduced the burden of human beings to a certain extent and changed the way o...

AI Technical Summary

Problems solved by technology

But at present, most researchers conduct research on innovations in algorithmic network models, and rarely conduct research on how to improve data utilization (a large amount of unlabeled data) and improve the utilization of error samples.

Images