Multi-modal small target detection method and system based on remote sensing image

Abstract

The invention discloses a multi-modal small target detection method and system based on a remote sensing image, and the method comprises the steps: collecting data containing RGB images and infrared images, and making a data set; performing morphological transformation on the RGB image and the infrared image; constructing a multi-modal small target detection network model through a shallow feature network, a balance fusion network, a deep feature network, a spatial pyramid layer, a path aggregation network and a multi-scale head network; designing a loss function based on a multi-scale head network, and training and testing a multi-modal small target detection network model through the data set; and performing small target detection on a to-be-detected multi-modal remote sensing image through the tested multi-modal small target detection network model. The method can be suitable for small targets, and complementarity of the RGB image and the infrared image is fully utilized. In the multi-modal data based on the remote sensing image, the method has an accurate detection effect.

Description

technical field [0001] The invention belongs to the technical field of target detection, in particular to a remote sensing image-based multi-modal vehicle small target detection method, system, device and storage medium. Background technique [0002] On the basis of real-time, dynamic and macro characteristics, remote sensing images provide a new detection method for military reconnaissance, geological disaster investigation and treatment. In recent years, with the rapid development of satellite remote sensing technology and deep convolutional neural network technology, remote sensing image target detection has important applications in military, intelligence, business, economics, planning and other fields. [0003] Some early work was based on traditional machine learning methods for detection, and its detection performance is very limited. With the development, the deep convolutional neural network occupies a dominant position in the target detection task, and the target ...

AI Technical Summary

Problems solved by technology

The two methods have advantages in speed and accuracy in general target detection tasks, but for remote sensing data collected by satellites, remote sensors, UAVs and other equipment, the effect of these algorithms is due to the small size of the remote sensing data target and is easily affected by light. , weather and other factors often fail to meet expectations, so there are certain differences from general target detection tasks, which brings many challenges to detection tasks

Although there have been a lot of remote sensing image target detection work, most of them are only aimed at the target detection of RGB images.

Images