Remote sensing image marine ship identification system and method based on improved YOLOv4 algorithm

Abstract

The invention discloses a remote sensing image marine ship identification system and method based on an improved YOLOv4 algorithm. The method comprises the steps of collecting satellite remote sensing images of a sea surface scene shot or collected in the past; performing type labeling on the preprocessed picture by using data labeling software; segmenting a ship in the remote sensing image from a surrounding environment phase to eliminate image noise; obtaining an estimated value of an anchor box of the YOLO algorithm; generating a YOLOv4 framework; generating a detection frame of the YOLOv4; setting a threshold value of the candidate box, and finally obtaining a prediction box; calculating three loss functions and minimizing the total value of the three loss functions to obtain an improved YOLOv4 neural network after training; and inputting the pictures in the test subsets into the trained and improved YOLOv4 network to obtain the target category, the specific position of the target in the pictures and the width and height of the target so as to complete target detection. The sea surface ship target can be rapidly detected and automatically identified, and the ship identification probability and accuracy are high.

Description

technical field [0001] The invention belongs to the technical field of ship target detection, and relates to a remote sensing image marine ship recognition system based on an improved YOLOv4 algorithm and a method thereof. Background technique [0002] The research on the identification of sea surface warship targets is a key technology in both military and civilian aspects, and has engineering application value. In terms of military use, it can be used for real-time automatic reconnaissance, laying the foundation for naval battlefield threat estimation and situation assessment. At present, the intelligence level of the photoelectric reconnaissance system is far lower than that of civilian electronic equipment. In the photoelectric reconnaissance process, the operator still relies on the operator to manually interpret the ship target according to the displayed image. The interpretation speed is slow and is easily affected by subjective factors. The use of target detection an...

AI Technical Summary

Problems solved by technology

[0004] In the Chinese patent application of CN201810457334.X, the SAR radar screen display is not intuitive enough to directly judge the surrounding environment, or the cold wave weather is affected by wind and waves, and it is difficult for ordinary monitoring equipment to identify key targets in harsh environments; in CN201911156854 In the Chinese patent of .8, the two-stage detection model represented by Faster-RCNN needs to first generate a candidate frame and then identify and detect it separately, and its calculation speed is slow, which is not conducive to practical application

In addition, in ocean pictures, satellite pictures and camera pictures are often troubled by problems such as foggy weather, motion blur, and camera lens pollution, resulting in blurred shooting results, which cause great interference to target recognition, and the accuracy of target recognition is even higher. Low

In the Chinese patent application of CN109255286B, although the YOLO series algorithms have achieved good results in real-time target detection and recognition, the YOLO series algorithms are not effective in recognizing remote sensing images.

Remote sensing images are different from natural images. Due to the long shooting distance in remote sensing images, ship targets are mostly small in size and occupy very few pixels in the entire image. If the IOU value (IOU=0.5) used to identify natural When remote sensing images, there will be missed detection

Images