Aircraft object identification method based on blur-invariant convolutional neural network

Abstract

The invention discloses an aircraft object identification method based on a blur-invariant convolutional neural network (BICNN), comprising the following steps: 1, establishing a remote sensing aircraft data set; 2, setting the number of layers of the BICNN; 3, improving an objective function; 4, set the number of fine-tuning layers and loading parameter; 5, calculating a forward propagation error; and 6, updating an weighty adjustment amount according to the error; and 7, testing the BICNN. The method can directly recognize a blur image, has blur invariance, a simple process and a high recognition rate, and has wide application scenarios.

Description

technical field [0001] The invention relates to an aircraft target recognition method, in particular to an aircraft target recognition method based on a fuzzy invariant convolutional neural network. Background technique [0002] As an important goal of air surveillance, aircraft targets are hot issues in current research to realize automatic identification of aircraft targets and models and to obtain more information. There is uncertainty and inaccuracy in the extraction of features, which reduces the accuracy of aircraft target recognition; [0003] Considering the influence of image blurring, there are two types of recognition methods for traditional blurred images. One is to deblur the image first, and then recognize the clear image through deblurring parameters. However, the same set of deblurring parameters cannot be used for different types of blurred images. better restoration; the other is to extract fuzzy invariant description sub-features for images, but these sub...

AI Technical Summary

Problems solved by technology

[0002] As an important goal of air surveillance, aircraft targets are hot issues in current research to realize automatic identification of aircraft targets and models and to obtain more information. There is uncertainty and inaccuracy in the extraction of features, which reduces the accuracy of aircraft target recognition;

[0003] Considering the influence of image blurring, there are two types of recognition methods for traditional blurred images. One is to deblur the image first, and then recognize the clear image through deblurring parameters. However, the same set of deblurring parameters cannot be used for different types of blurred images. better restoration; the other is to extract fuzzy invariant description sub-features for images, but these sub-features are not highly expressive for contour information, so they have certain limitations in application. In summary, traditional The fuzzy target recognition algorithm of the present invention has the problem that the recognition process is complex and the application environment is relatively single. To solve this problem, the present invention proposes a fuzzy target recognition algorithm based on a convolutional neural network;

[0004] In recent years, convolutional neural network, as a multi-layer feed-forward deep learning model, can directly use images as input and automatically learn features through multi-hidden layer structures, so there have been many successful applications in image-related tasks. , although the application and research of methods based on convolutional neural networks in aircraft recognition are gradually increasing, the research on robust recognition for target fuzzy invariance is still in the preliminary stage, and there is still a large research space. Based on this, the present invention proposes a method based on Blur-Invariant Convolutional Neural Network BICNN (Blur-Invariant Convolutional Neural Network), through the BICNN network can accurately identify fuzzy targets

Images