High-resolution remote sensing image inclined ship target detection method based on position sensing

Abstract

The invention relates to a high-resolution remote sensing image inclined ship target detection method based on position sensing. The method comprises the following steps: extracting a multi-scale depth feature map by using a UNet-like multi-scale convolution network; directly predicting the possibility score that each anchor point has a ship and the position of the anchor point in a prediction boxthrough a rotating box regression model based on the anchor points by utilizing the extracted depth feature map; and correcting the anchor point score by using the position sensing score correction model. According to the method, the UNet-like convolutional neural network is utilized to extract both deep semantic features and shallow detail features, the model positioning precision is improved while the classification precision is ensured, and the detection performance of small-size ships is improved. And meanwhile, the candidate box score is further corrected by utilizing the position sensing score correction model, so that the candidate box positioning precision is improved.

Description

technical field [0001] The invention relates to a method for detecting ship targets in remote sensing images, in particular to a method for detecting obliquely densely arranged ship targets from high-resolution remote sensing images. Background technique [0002] In the past few decades, ship detection has been a hot topic in the field of remote sensing, which plays an important role in promoting the development of national defense construction, port management and cargo transportation. The traditional ship detection algorithm realizes ship detection by extracting and identifying the shape and texture features of ships. This kind of method is simple, easy to implement, and has strong interpretability, but its extracted features are mostly shallow information, and it is necessary to design a method suitable for all ships. Manual features are more difficult. [0003] Convolutional neural networks have made significant progress in the field of object detection. However, ships...

AI Technical Summary

Problems solved by technology

However, a large number of rotating anchors increases the difficulty of object classification and generates more false alarms

"Rotated Region Based Fully Convolutional Network for Ship Detection" (IGRSS.IEEE,2018:673-676.) proposed a rotating ship detection method based on a fully convolutional neural network. This method has a faster detection speed, but due to positive The proportion of samples and negative samples is unbalanced, and the accuracy of detection in complex backgrounds is low

Images