Method, device and system for image data classification and classification model establishment

Abstract

The present invention relates to a method, device and system for classifying image data and establishing a classification model, wherein the method for establishing a neural network classification model includes: acquiring at least one frame of image data containing a target object; Key areas, each key area in several key areas is an area whose probability of at least containing part of the target object meets the predetermined requirements; at least one key area is selected from several key areas as the data input of the neural network, and the neural network is trained ; Adjust the structural parameters of the neural network according to the training results output by the neural network to establish a neural network classification model. The invention can effectively remove noise and improve the accuracy of image data recognition / classification.

Description

technical field [0001] The invention relates to the field of video image processing, in particular to a method, device and system for classifying image data and establishing a classification model. Background technique [0002] Multiple instance learning is an important weakly supervised learning algorithm. In multiple-instance learning, samples are given in the form of bags. Each package is composed of multiple examples (instance). For a binary classification problem, a positive bag contains at least one positive example, while a negative bag contains only negative examples. The task of multi-instance learning is to simultaneously infer labels for each example and learn a classifier. [0003] In image classification, an image might be annotated with a label of some kind of object, such as "dog". But the region in the image associated with the label (pixels occupied by the dog) only occupies a small part of the image, and most of the regions may not be associated with th...

AI Technical Summary

Problems solved by technology

However, the area of ​​the image related to the label (pixels occupied by the dog) only occupies a small part of the image, and most of the area may not be related to the label.

When a deep neural network is used for video classification, a sub-region (volume) is randomly selected from the spatio-temporal three-dimensional space of the video as the sampling area (for example, the time span is 10 frames), and the sampling area is allowed to inherit the video label. This can lead to noisy data

Images