Trusted target detection method fusing subjective logic and uncertainty distribution modeling

Abstract

The invention discloses a credible target detection method fusing subjective logic and uncertainty distribution modeling. The method comprises the steps that 1, image data are collected, the image data are preprocessed, target labeling is conducted on preprocessed images, and a first data set is constructed; step 2, constructing a credible target detection model; 3, training the credible target detection model to obtain a trained credible target detection model; and step 4, inputting a test image in the first data set into the trained credible target detection model, and obtaining target positioning information, a target category identification result and an overall credibility evaluation result of the target category identification result in the image. Compared with an existing target detection method, the method has the advantages that the recall ratio of target detection is increased, a target detection result is given, credibility evaluation is carried out on a target category recognition result, and a basis is provided for subsequent decision making of a user.

Description

technical field [0001] The invention relates to the technical neighborhood of target detection, in particular to a credible target detection method integrating subjective logic and uncertainty distribution modeling. Background technique [0002] Image target detection technology is widely used in artificial intelligence, unmanned driving, and smart medical care. The field of machine learning has grown tremendously with the advent of large visual image datasets and deep learning techniques. Today, deterministic neural networks have been proved to be extremely efficient image detection classifiers, and have shown amazing results in related work, and the most representative of them can be divided into R-CNN (Region-based Convolutional NeuralNetworks) as the representative of the two-stage target detection network and the YOLO (YouOnly Look Once) single-stage target detection network that transforms the target detection problem into a regression problem. At the same time, than...

AI Technical Summary

Problems solved by technology

[0005] In view of the above-mentioned various problems, the deterministic neural network cannot give a result with credibility evaluation, so it is necessary to invent a credible target detection method that integrates subjective logic and performs uncertainty distribution modeling

Images