SAR image target detection and recognition integration method

Abstract

The invention relates to a radar remote sensing application technology analyzing radar observation information through images, in particular to an SAR image target detection and recognition integration method based on a convolutional neural network (CNN). According to the method, the convolutional neural network is utilized to automatically mine and select target features, shallow features and deep features are fused, a detection task and a recognition task of an SAR target can be completed at the same time, and SAR image target detection and recognition integration is realized. Compared with other SAR target detection and recognition algorithms, the method has higher detection and recognition efficiency and higher applicability.

Description

technical field [0001] The invention relates to radar remote sensing application technology, which uses images to analyze radar observation information, in particular to an integrated method for detection and recognition of SAR image targets, which is based on a convolutional neural network (CNN). Background technique [0002] The detection and recognition of Synthetic Aperture Radar (SAR) images is one of the important research contents in the field of radar remote sensing applications. Carrying out the research on SAR image target detection and recognition is of great significance to promote the scientific development of radar remote sensing application technology. [0003] Compared with optical images, the biggest feature of SAR images is the influence of coherent speckle noise. Its existence makes SAR images show low signal-to-noise ratio, so many standard optical image target detection and recognition algorithms are difficult to be satisfied when applied to SAR images. ...

AI Technical Summary

Problems solved by technology

The above-mentioned first idea based on contrast makes use of the most obvious features in SAR images, but there are still many unfavorable factors, such as the difficulty in establishing a clutter statistical model, the difficulty in adaptively selecting the detection threshold of the detector, and the slow running speed of the algorithm. Prior knowledge is extremely complex to train, etc.

The reason why the second idea based on other features of the image is difficult to be further applied is that manual feature extraction is too difficult and the process is too cumbersome

The third problem with the idea based on complex image features is that most of the existing algorithms of this type have low-frequency SAR detection characteristics, and cannot be well applied to high-frequency SAR image detection tasks.

The disadvantage of the first idea is that the template library of the target increases with the increase of the target type, which requires a large amount of storage space and affects the recognition speed and recognition accuracy.

The second way of thinking generally requires high image quality and high-fidelity CAD modeling technology. When the observation conditions change and the image does not match the model, the recognition effect will be greatly affected.

[0007] However, most of these methods will perform noise reduction processing before detection and recognition, and at the same time separate the target detection and target recognition tasks. There is a certain image processing process in the detection and recognition process, so there is no one that can realize the integration of detection and recognition. Methods

Such a process will introduce certain errors, which will greatly affect the recognition efficiency.

At the same time, the above-mentioned target detection and recognition methods still rely on the tedious manual exploration, selection and extraction of target features, and do not make full use of target shallow features and deep features for feature combination to improve detection and recognition efficiency. To achieve the best detection and recognition effect A small improvement will cause a sharp increase in the complexity of the algorithm, sacrificing a lot of time cost

Images