Optical Remote Sensing Image Ship Rotating Object Detection Based on Improved YOLOv7
Aiming at the characteristics of large scale change and arbitrary direction of ship targets in the complex background of remote sensing images,an improved ship rotating target detection algo-rithm based on YOLOv7 was proposed.By incorporating the Global Attention Mechanism(GAM)in-to the backbone network,the model was guaranteed to pay more attention to the ship objects in the complex background.By adding an adaptive spatial feature fusion module(ASFF)to the neck,the problem of inconsistent feature information in the feature pyramid was eliminated.Aiming at the char-acteristics of slender ship scale and arbitrary direction,by adding rotating frame detection represented by Gaussian modeling,the positioning accuracy was improved and the ship target direction information was retained.The results show that the improved algorithm achieves 90.73％detection accuracy in HRSC2016 data set while ensuring the detection speed,and the detection effect is also improved.
万方数据
维普
