Multi-mode small target detection method based on knowledge distillation

Abstract

The invention discloses a multi-mode small target detection method based on knowledge distillation, and the method comprises the steps of constructing a visible light-multispectral image data intrinsic knowledge migration model, carrying out the feature refining through a spatial-spectral combined non-local feature pyramid visual attention structure, and refining the multi-spectral and spatial-spectral combined features through a knowledge distillation model based on spatial-spectral combined feature migration to obtain a high-precision and high-efficiency deep neural network. According to the invention, target intrinsic knowledge can be migrated by using visible light big data to solve the problem of insufficient multispectral data samples, the detection and recognition robustness is improved through non-local attention, the number of neural network parameters after knowledge distillation is simplified, the computing resource overhead is reduced, and lightweight operation can be realized.

Description

technical field [0001] The invention relates to multi-mode small target detection and recognition, in particular to a multi-mode small target detection method based on knowledge distillation. Background technique [0002] Through multi-mode imaging, there are both spatial information and spectral information; the comprehensive use of these information for target recognition has greatly improved the accuracy of target recognition compared to using visible light and infrared data alone. This technology has a wide range of application requirements in many fields, including searching and rescuing targets such as people, ships, and floating objects at sea, and detecting and identifying low-slow and small targets such as aircraft, drones, and birds in airports and urban security systems; at the same time, it is also It plays an important role in key air defense, aviation detection, forest fire prevention and other scenarios. [0003] At present, there are some methods for multi-m...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a multi-mode small target detection method based on knowledge distillation, which can not only solve the problem that the number of multi-spectral data samples is insufficient to support high-precision neural network model training in a specific environment, but also It has the ability to improve the accuracy and robustness of the model under the interference of complex environments and noises. At the same time, the model detection accuracy is higher and the training efficiency is higher.

Images