Remote sensing target detection method of dual attention mechanism

Abstract

The invention discloses a remote sensing target detection method of a double attention mechanism, and belongs to the field of image processing in computer vision. The method comprises the following steps: 1, inputting a multi-scale remote sensing aerial image; 2, segmenting the input picture into 512 * 512 small pictures; 3, inputting the small picture into a network, and generating a feature mapthrough a receptive field module; 4, constructing a double attention mechanism convolutional neural network as a matching function, optimizing network parameters, and redefining a feature map generated through the convolutional neural network; 5, realizing classification and bounding box regression of the targets through subsequent operation of the network; and 6, obtaining a regression box with the highest score by inhibiting the non-maximum values of the plurality of regression coordinate boxes, wherein the box is an interested target position in the remote sensing aerial image. According tothe invention, the target in the remote sensing aerial photo can be effectively detected, and the accuracy and generalization of the target detection network are improved.

Description

technical field [0001] The invention relates to a remote sensing target detection method with a dual attention mechanism, which belongs to the field of computer vision image processing. Background technique [0002] Target detection in remote sensing images has a huge demand in military and civilian applications. For example, when ships are sailing, it is necessary to use satellite imaging technology to collect images and quickly detect ships in remote sensing images to help sailing ships stay away from danger and ensure safe navigation on sea and river surfaces. The military has an important demand for the rapid detection of aircraft in remote sensing images. In the process of real-time detection of aircraft in remote sensing images, artificial intelligence of spy satellites can be realized. This can be a great advantage in modern warfare. And the method of remote sensing target detection can be used to detect the approximate position of clouds in the atmosphere in real t...

AI Technical Summary

Problems solved by technology

However, there are some problems in the single-stage target detection algorithm. The skeleton network used in the network is a shallow convolutional neural network, and the network model designed in this way has weak representation ability.

Especially when the target of interest in remote sensing aerial images often faces a complex background, the target of interest is doped in a background with a lot of noise, and the target of interest faces the problem of large changes in the target scale, due to the representation ability of the network Insufficient and weak discrimination makes the target detection network often miss detection and false detection, which reduces the detection effectiveness and detection accuracy of the model

Images