SAR image ship target rapid detection method based on lightweight convolutional neural network

Abstract

The invention provides an SAR image ship target rapid detection method based on a lightweight convolutional neural network, and the method comprises the steps: inputting all feature maps outputted bya convolution layer of a cascading unit of cascading convolution modules into a feature aggregation unit for feature aggregation, and inputting an obtained aggregation feature map into a next cascading convolution module; performing feature aggregation on feature maps output by the two channels of a dual-channel convolution module and then inputting the result into the next dual-channel convolution module. Therefore, the utilization rate of the feature map generated by each convolution layer can be improved, the parameter quantity can be reduced, and the accuracy of a target detection system can be ensured under the condition of reducing the requirement on the data quantity of an original SAR image.

Description

technical field [0001] The invention belongs to the field of radar remote sensing technology and computer vision technology, and in particular relates to a method for quickly detecting ship targets in SAR images based on a lightweight convolutional neural network. Background technique [0002] SAR is the acronym for Synthetic Aperture Radar, which refers to synthetic aperture radar, which is an active microwave remote sensing imaging radar that can work all day and all day long. It has extensive and important applications in civilian and civilian fields, and plays an irreplaceable role in other remote sensing methods such as optics and infrared. As a large maritime country with tens of thousands of kilometers of coastline and millions of square kilometers of ocean land area, in order to safeguard national territorial sea sovereignty and legitimate rights and interests from illegal infringement, the rapid detection of ship targets at sea based on SAR images is necessary to ti...

AI Technical Summary

Problems solved by technology

However, there are great differences between the source domain and SAR images in terms of statistics, vision, and transformation domain features, which limits the feature representation of ship targets in SAR images by CNN; ", the detection speed is slow, and it cannot be integrated into portable and embedded devices with limited computing resources in practical applications

For this reason, abandoning the transfer learning technology and starting training from a network with parameters randomly initialized (that is, training from zero) can effectively solve this problem; at the same time, considering the cost of obtaining and making a large-scale and accurately labeled ship target SAR image dataset It is too high, the cycle is long, and the difficulty is high. Therefore, it is a solution under the existing conditions to design a lightweight and zero-trained SAR image ship target detection method with a small amount of parameters under the premise of ensuring a certain detection accuracy. An effective approach to the above problems

Images