Target detection method for remote sensing image

• Main Authors: WANG NANA, ZHAO LIANCHEN, PENG YIZHUN
• Format: Patent
• Language: Chinese; English
• Published: 27.06.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

A multi-scale feature fusion remote sensing image target detection method based on non-local feature enhancement comprises the steps that firstly, an overall framework is based on an improved Faster RCNN network, a backbone network adopts a ResNet101 + FPN structure to replace an original VGG network, and a bottom-up enhanced feature pyramid is constructed; for multi-scale features generated by FPN, three appropriate outputs are selected and input into an improved RPN network, the RPN network is added into a non-local feature enhancement module, multi-scale feature fusion is applied to generate a more appropriate recommendation area, and the speed is correspondingly increased. The method solves the problems that existing remote sensing image target detection precision is not high and missing detection and false detection occur, and the detection effect is good. Meanwhile, the invention provides a detection system, so that the detection process is clearer, and the flexibility is higher. 一种基于非局部特征增强的多尺度特征融合的遥感图