An image analysis method with high recognition rate based on depth learning

Abstract

The invention discloses an image analysis method with high recognition rate based on depth learning which is used for generating an image analysis system. The method comprises that follow steps of: collecting data, establishing a common object data set, classifying the common object data set into different classified common samples according to an object class, and dividing the classified common samples into different specific common samples; object labeling is performed on the sub-samples of the specific common samples; aiming at the specific object, the data set of the specific object is setup, the data set of the specific object is classified into different classification specific samples according to the object category, and the specific samples of each classification are divided intodifferent specific samples concretely; object labeling of sub-samples of a specific sample. The object-specific model is obtained by model training for each specific sample combined with the target detection algorithm. The system generated by the method is suitable for special fields such as public security, judicature and the like, and supports user-defined expansion, and the user can establishthe object-specific model according to his own requirements, and the image recognition rate is high.

Description

technical field [0001] The invention relates to the field of image recognition, in particular to an image analysis method based on deep learning and high recognition rate. Background technique [0002] In the field of electronic data forensics such as public security and justice, there are a large number of image recognition needs. At present, most of them are classified and screened manually. It often takes hours or even days for a single evidence source sample, which costs a lot of manpower and time; With the development of the Internet, the amount of data has expanded significantly, and the manual processing method has become increasingly unable to keep up with the pace of cases. Most of the existing intelligent identification and classification schemes are the identification of some common object categories, and due to the particularity of the fields of public security and justice, it is necessary to identify a large number of sensitive content, such as: murder weapons, ...

AI Technical Summary

Problems solved by technology

The field of forensics involves many complex scenes and synthetic images. Most of the current intelligent recognition model training schemes are simply classified by features, and the recognition rate for more complex images is very low.

And currently there is no supported intelligent recognition system user-defined extension function

Images