A target detection method based on generative adversarial network

Abstract

The invention provides a target detection method based on a generative confrontation network, which includes designing a generator, generating various samples according to category labels, designing an agent, detecting the data of the generator, providing false true values, and applying the data generated by the agent to Training of target detectors, designing target detectors, judging whether the generated data is conducive to improving the accuracy of target detection, designing adversaries, in the training phase, judging whether the data comes from real data or generated data, the generator and the discriminator are trained alternately, In the testing phase, the data to be detected is directly input into the target detector to obtain the detection result. The combination of the samples generated by the generation network of the invention and the real samples can enrich the training data, improve the detection accuracy, the target detection network provides feedback to the generation network, so that the generated samples are more realistic, and the data generated by the agent is directly applied to the training of the target detector. There is no need to spend a lot of manpower and material resources for labeling, and the present invention has a simple structure and is easy to deploy.

Description

technical field [0001] The invention belongs to the field of target detection methods, in particular to a target detection method based on a generative confrontation network. Background technique [0002] In recent years, deep learning has developed rapidly, and target detection algorithms based on deep learning have achieved good results, but there are still problems to be solved. First, target detection algorithms based on deep learning require a large number of labeled samples, and deep learning is often prone to overfitting on small data sets. According to research, a rough rule of thumb is that supervised deep learning algorithms will generally achieve acceptable performance given about 5,000 labeled samples per class, when a dataset of at least 10 million labeled samples is used for training. will meet or exceed human performance. In addition, the labeling of the dataset will also consume a lot of manpower and material resources. Therefore, how to use a large number...

AI Technical Summary

Problems solved by technology

Secondly, most of the target detection research at this stage relies on deeper networks to improve accuracy, but this will bring more complex calculations and more memory requirements, consume a lot of hardware, and be difficult to deploy

Generating

Images