Method for detecting small target of high-resolution image of any scale

Abstract

The invention discloses an arbitrary-scale high-resolution image small target detection method. The method comprises the following steps: obtaining a marked and cleaned target data set, and dividing the target data set into a training set and a test set; calculating a preset anchor box for the data set through an optimization clustering algorithm; designing a convolutional neural network, obtaining a feature map through feature extraction, adding patches to the prediction network to enable the feature map obtained through up-sampling and the feature map corresponding to the feature extractionlayer to be consistent in dimension, and performing multi-scale detection on the feature maps of different scales; training the data set by using a convolutional neural network, and obtaining a neuralnetwork model with small target detection capability after the performance evaluation indexes converge; and detecting the test data set to obtain a target category and position coordinates. The method is suitable for the image input network of any size, avoids the loss of small target features on the feature map after being reduced to the fixed-size input network or the loss of edge target context information caused by image cutting, facilitates the detection of a small target of a high-resolution image, and is wide in applicability.

Description

technical field [0001] The invention belongs to the technical field of deep learning image processing and small target detection of remote sensing high-resolution images, and in particular relates to a method for detecting small targets of high-resolution images of any scale. Background technique [0002] With the development and application of satellite remote sensing technology and computer vision technology, target detection in optical remote sensing images is of great significance in civilian and military applications. In civilian applications, high-precision target detection helps to assist traffic management and planning; in military applications, high-precision target detection contributes to accurate intelligence and reconnaissance, accurately locks hostile target intrusions and hazards, and maintains national security. However, high resolution and small targets are difficult and one of the most prominent problems in remote sensing image target detection. [0003] T...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to provide a small target detection method of high-resolution images of any scale, which overcomes the fixed input image size of the existing deep learning network. The lack of missed detection of small targets improves the accuracy of small target recognition in the case of large-size images

Images