Method, system and terminal for detecting and identifying ship target in remote sensing SAR (Synthetic Aperture Radar image

Abstract

The invention belongs to the technical field of SAR ship target detection and identification, and discloses a ship target detection and identification method, system and terminal in a remote sensing SAR image. The method comprises: carrying out data preprocessing, i.e., carrying out triple mean stretching processing on a shot SAR image; cutting a test image through a sliding window, i.e. dividing the preprocessed SAR image into image blocks with a preset size through a sliding window mode; predicting the position and the size of a target by using a detection network, i.e. performing binary target detection on each small image block by using the pre-trained detection network; performing false alarm filtering by using an identification network, i.e. identifying each detection result by using a pre-trained identification network, and filtering detected false alarms; and taking the identified type and position as a final result. The small sample transfer learning method based on parameter regularization is adopted, the requirement for the number of each class of training samples is low, the actual application conditions of the SAR field are met, and the invention has the actual application value.

Description

technical field [0001] The invention belongs to the technical field of SAR ship target detection and recognition, and in particular relates to a ship target detection and recognition method, system and terminal in remote sensing SAR images. Background technique [0002] At present, because it can penetrate clouds, fog, rain, smoke, etc., and can work around the clock, SAR, as an effective means of remote sensing reconnaissance and surveillance, has been equipped on different types and multiple levels of platforms. It is a very important imaging perception component of various types of military equipment in service. It can be used for military target reconnaissance, weapon precision guidance and battlefield situation awareness, and provides discrimination information such as target attributes and categories. It is unique and irreplaceable. Therefore, with the increasing demand for environmental perception of battlefield research and judgment systems and military equipment, ta...

AI Technical Summary

Problems solved by technology

[0006] (1) Compared with optical remote sensing images, SAR image training data is difficult to obtain and label in large quantities, and it is difficult to meet the training requirements of deep networks

[0007] (2) At present, the false alarm rate of the algorithm for ship target detection in remote sensing SAR images is high, which is difficult to meet the actual application requirements

[0008] (3) The existing level of automation and intelligence for remote sensing SAR images cannot match the existing target detection capabilities, and it is an urgent problem to improve combat effectiveness such as battlefield reconnaissance and situational awareness when developing automatic target detection and recognition

[0010] (1) The existing deep network-based target detection and recognition framework requires a large number of training samples due to its complex structure, and it is necessary to study the model structure and training method that can successfully train the deep network with a small amount of training data

[0011] (2) The background information of remote sensing SAR images is complex, and there are many background information similar to the target, resulting in a large number of false alarms in the results of the current common detection and recognition algorithms

[0015] (2) A higher false alarm will lead to a decrease in the credibility of the algorithm in actual use, so reducing the false alarm rate of the algorithm is of great significance to improve the actual application value of the algorithm

Images