Deep learning method for ship detection in high-resolution visible remote sensing images

Abstract

The invention provides a deep learning method for ship detection in high-resolution visible remote sensing images, which comprises the following steps: firstly, reading and preprocessing image data; secondly, extracting the features of the whole image; thirdly, screening out target candidate regions after extracting the abstract features of the image in the convolution layer; fourthly, cutting outthe feature blocks of each target candidate region on the feature map corresponding to the whole image, and using the pooling layer in the region of interest to normalize the sizes of the feature blocks; fifthly, sending the features to the full connection layer to get spatial transformation parameters, and sending the spatial transformation parameters and the features to the spatial transformation layer to get the features after deformation correction; and sixthly, carrying out classification and position correction again on the target candidate regions according to the corrected features. The robustness of the detection method to target rotation and other deformation is enhanced, and the detection effect of ship targets in high-resolution visible remote sensing images is improved. The method can be applied to the detection of ship targets in high-resolution visible remote sensing images, and has broad application prospects and values.

Description

(1) Technical field: [0001] The invention relates to a deep learning method for ship detection in high-resolution visible light remote sensing images based on faster R-CNN (faster Region-Convolutional Neural Network) and STN (Spatial Transformer Network) in deep learning, belonging to high-resolution remote sensing images Target detection technology field. (two) background technology: [0002] Remote sensing technology generally refers to a technical method that transmits or receives electromagnetic waves including light waves through sensors and other equipment on aircraft, satellites and other aircraft, so as to obtain and analyze the characteristics of ground targets. In recent years, with the upgrading of sensor equipment and the improvement of information processing level, the resolution level of remote sensing images has been greatly improved, forming a large number of high-resolution visible light remote sensing images with clear texture and rich details. Target dete...

AI Technical Summary

Problems solved by technology

[0005] Up to now, the most representative of the two-stage detection network is the faster R-CNN, but the faster R-CNN is a network designed for the purpose of detecting multiple types of targets in natural images, which cannot be well suited to remote sensing images. The characteristics of the ship target (such as the long and thin shape of the ship in the remote sensing image, multiple rotation angles, etc.)

Images