Ensemble classification-based violence and terrorism image labeling method

Abstract

The invention discloses a violent and terrorist image labeling method based on ensemble classification. The method comprises the following steps: normalizing a to-be-annotated image into a size of 224* 224, and then inputting the image into an image annotation integration network; mapping the extracted image features into a label semantic space by each sub-network in the integrated network to obtain a label probability vector of which the size of the to-be-labeled image is 1 * N; combining the label probability vectors output by the sub-networks into a matrix, and performing a series of operations on the matrix and the weight allocation matrix to obtain a label probability vector of which the final size of the to-be-labeled image is 1 * N; and setting a threshold value for the calculatedlabel probability vector, wherein all labels greater than the threshold value are final labeling results of the to-be-labeled image. The annotation method described by the invention has the advantagesof short network training time, high annotation accuracy, strong stability and the like, compared with traditional machine learning, the annotation accuracy and recall rate are greatly improved, andthe annotation method has a certain practical value for the specific field of terrorist-related information.

Description

technical field [0001] The present invention designs a method for labeling images of violence and terrorism based on integrated classification, which relates to the technical fields of deep learning and computer vision. Background technique [0002] With the rapid development of Internet social platforms and the popularization of image acquisition devices such as mobile phones and digital cameras, the image and video data that people can access every day has shown explosive growth. While massive image data brings convenience to people's daily life, some violent and terrorist images also have a negative impact on social harmony and the healthy growth of young people. How to effectively manage these data has become an urgent problem to be solved. Automatic image annotation technology has gradually become one of the key technologies in the field of image analysis and application because of its feature of automatically adding text feature information reflecting its content to i...

AI Technical Summary

Problems solved by technology

[0004] The traditional method has made some progress in the field of image annotation, but because of the need to manually select features, resulting in information loss, resulting in insufficient annotation accuracy and low recall rate; although the deep learning model has made relatively high achievements in the field of image recognition and classification , but most of them are improvements for the network itself or for single-label learning, while there are fewer applications and improvements for image annotations that belong to multi-label learning

The imbalance of label categories in the multi-label database will also lead to a reduction in the quality of the trained model annotations

Images