基于FasterR-CNN的小样本小目标检测研究

打开文本图片集

中图分类号：TP183 文献标识码：A 文章编号：2096-4706（2025）15-0067-06

Research on Small Sample and Small Target Detection Based on Faster R-CNN

SUN Yanyi， YANG Bowen， SI Make， MU Weimin， LEI Liang （Shanghai InstituteofMechanical and Electrical Engineering，Shanghai 2O11O9，China）

Abstract：Due to the existence of certain dependence problems in the currnt database，there is a certain mismatch betweenthe training test setand thereal situation.Based on TransferLeaming，this paper takes the improved TransferLearming asthe method，ombedwithasterR-CNNalgoritm，especiallyfortedetectionandcasificationofsalltargetsinteai andstrives tosolvetheproblemofmisidentificationofsmalltargetsintheairThe modelistrainedontherealandsimulation databaseaccordingtothetheoretical model，soas tosolve thedificultiessuchasthecomplexityoftargetatributes，the complexityofdecoysorenvironmental interference.Thispapercompares theinfuenceofthemethodwithoutusingtheimproved network andthecommonlyused TransferLearning methodontheexperimental results inthe detectionof smalltargets inthe air from heaspectsof networkandsampleatributes.The experimentalresults showthatthe detectionaccracyanddetection effciencyoftheaerialsmaltarget detection methodbasedonthe improvednetwork are improvedcomparedwithotherTransfer Learning networks.

KeyWords：Deep Leaming;MachineLearning;TransferLearning;FasterR-CNNalgorithm;TargetDetection;Neural Vetwork; small target

0 引言

目标侦察的主要任务是探测复杂环境下的空中小目标并识别目标类型与属性，精确自主制导的主要任务是跟踪目标、抵抗干扰并在末端制导阶段定位到目标的重点部位。（剩余7812字）