A ship detection method for optical remote sensing images based on the improved yolo V2 model

Abstract

The invention discloses an optical remote sensing image ship detection method based on the improved YOLO V2 model. First, the remote sensing image is preprocessed, and then a single YOLO V2 network model is used to realize the process of ship target feature extraction, detection, and target positioning in the remote sensing image, so as to overcome the shortcomings of artificial feature extraction and dig deep into the information contained in the data; at the same time, considering the ship The difference in losses caused by misjudgment of false alarms, false alarms as ships, and one type of false alarms as another type of false alarms is weighted and corrected for the category misjudgment loss of the YOLO V2 model, which strengthens the network’s ability to detect ships. The feature learning of ship target samples reduces the number of training algebras on the basis of ensuring the detection performance of ships; the misjudgment between false alarms is weighted with a factor of 0, which reduces the accuracy requirements for false alarm calibration. Affects the convergence of the loss function.

Description

technical field [0001] The invention belongs to the field of target detection and relates to a ship target detection method based on an improved YOLO V2 model of an optical remote sensing image. Background technique [0002] Ship detection plays a prominent role in national marine security, marine management, monitoring illegal fishing, etc. With the development of remote sensing technology, the resolution of optical remote sensing images has been continuously improved, and the amount of information has become more and more abundant. Ship target detection based on optical remote sensing images has become a major research hotspot today. Most of the usual ship target detection algorithms first extract the suspected target area, then perform manual feature extraction based on expert knowledge, and finally use machine learning methods to classify and identify targets to obtain the final detection results. [0003] In recent years, with the blowout growth of data volume and the...

AI Technical Summary

Problems solved by technology

[0005] In order to use a single YOLO network model to realize the process of ship target feature extraction, detection, and target positioning, overcome the shortcomings of manual feature extraction and dig deep into the information contained in the data. The difference in loss caused by distinguishing a certain type of false alarm from another type of false alarm

Images