Optical remote sensing image multi-class target detection method based on cross-scale feature fusion

Abstract

The invention relates to an optical remote sensing image multi-class target detection method based on cross-scale feature fusion. The method comprises the steps: taking training data as the input of aconvolutional neural network to extract image features, obtaining a multi-scale feature map from the output of different convolutional layers, and adding an extrusion-excitation module at the topmostfeature to remodel the channel information of the top feature; performing cross-scale feature fusion operation on the obtained feature maps, training a region suggestion network on the multi-scale feature maps, obtaining a suggestion box for subsequent tasks from the trained region suggestion network, and sending the suggestion box to a classification network and a regression network for training; and finally, realizing accurate detection of multiple types of targets of the optical remote sensing image on the multi-scale feature map through post-processing operations such as non-maximum suppression and the like. By using the method of the invention, various types of targets can be detected from an optical remote sensing image under a complex background. The method has higher detection andidentification precision and higher speed.

Description

technical field [0001] The invention belongs to a multi-type target detection method based on optical remote sensing images, and relates to a multi-type target detection method in optical remote sensing images based on cross-scale feature fusion, which realizes cross-scale fusion of features and can be applied to multiple types of optical remote sensing images with complex backgrounds. Types of object detection tasks. Background technique [0002] With the rapid development of aerial remote sensing technology, it has become easier and easier to obtain a large amount of remote sensing data from high altitudes. At the same time, various tasks based on remote sensing images emerge in an endless stream, such as object detection, scene classification, data compression, etc. As an application of remote sensing image processing technology, object detection in complex background optical remote sensing images is a key technology in the field of image processing, and has always been ...

AI Technical Summary

Problems solved by technology

[0003] At present, the realization of target detection tasks mainly adopts the following two types of methods: one is a single-stage target detection method, and the more representative work is published by Ross Girshick et al. on "IEEE Conference on Computer Vision and Pattern Recognition 2016" "You Only Look Once: Unified, Real-Time Object Detection", this method regards the target detection task as a regression task. The advantage of this type of method is that the detection speed is fast, and the disadvantage is that the detection accuracy is comparable to the two-stage target detection method. The other is a two-stage target detection method. This method first generates a series of anchor boxes in the image. The length, width, ratio, and quantity of these anchor boxes can be set according to task requirements. Network (Region Proposal Network, RPN) to solve the binary classification problem that the target in the anchor point is foreground or background, and perform a rough regression on the original anchor frame coordinates, and then perform classification and regression tasks. The advantage of this type of method is to detect High accuracy, the disadvantage is that the speed is slow, and the time spent in the task inference process is slightly longer

However, remote sensing images are quite different from natural scene images. Due to different imaging platforms and imaging methods, objects in optical remote sensing images have different degrees of deformation, occlusion, scale change and direction diversity. For small targets, their characteristic information is often overwhelmed by complex surrounding scenes. Some categories of targets are arranged too densely, and some categories of targets have high similarity in color and appearance. These problems are all for optical remote sensing image target detection. task increased difficulty

Images