Unsupervised target detection method and system based on variational auto-encoder and Gaussian mixture model

Abstract

The invention discloses an unsupervised target detection method and a system based on a variational auto-encoder and a Gaussian mixture model, and the method comprises the steps: converting an input image into a feature map of H * W dimensions, namely H * W cells, through a backbone network, coding the feature map into a hidden variable of which the prior distribution accords with the Gaussian mixture model, and carrying out the image reconstruction through a decoder according to the hidden variable, and comparing the reconstructed image with the input image, and calculating a loss function, so that a neural network is trained, an encoder obtains information such as the category and the position of an object in the image, and unsupervised target detection is realized. According to the method, the space attention mechanism and the Gaussian mixture model are combined, end-to-end target detection and classification can be achieved, meanwhile, good performance is still achieved under the condition that a large number of objects exist, and good expansibility is achieved.

Description

technical field [0001] The present invention relates to a technology in the field of artificial intelligence, in particular to an unsupervised target detection method and system based on a variational autoencoder and a Gaussian mixture model. Background technique [0002] Current supervised learning still requires large labeled datasets, and this processing requires a lot of work, making useful datasets hard to come by. At the same time, the generality of the deep learning model obtained by supervised learning is poor, and the performance on different data sets may be attenuated. In contrast, the biggest feature of unsupervised learning is that it does not need to label the data, which greatly reduces the workload. At the same time, unsupervised learning is committed to obtaining a general model and improving the versatility of the model. In the field of unsupervised learning, the variational autoencoder is a very important framework, although the AIR (Attend, Infer, Repea...

AI Technical Summary

Problems solved by technology

[0003] The present invention aims at the problem of insufficient detection accuracy of the existing unsupervised learning target detection for classification and multi-object scenes, and the existing target detection framework based on variational autoencoders is difficult to deal with scenes with a large number of objects and is not sensitive to object type information In order to solve the defects, an unsupervised object detection method and system based on variational autoencoder and Gaussian mixture model is proposed, which combines the spatial attention mechanism and Gaussian mixture model, which can not only achieve end-to-end object detection and classification, but also in the presence of In the case of a large number of objects, it still has good performance and has good scalability

Images