噪聲偽標(biāo)簽容忍的半監(jiān)督SAR目標(biāo)識別

打開文本圖片集

中圖分類號： TP391 文獻(xiàn)標(biāo)志碼：A DOI：10.12305/j.issn.1001-506X.2025.06.08

Abstract：To address thechallenge of limited recognition accuracy due to noise pseudo-labels in semisupervised synthetic aperture radar（SAR）automatic target recognition（ATR）with scarce labeled training samples，a noise pseudo-label tolerant semi-supervised SAR ATR method is proposed.The proposed method includestwo stages.In the first stage，high-reliability pseudo-labelsare generatedand selected by combining residual network（ResNet）and multi-clasifier fusion，soas toenrich the labeled training dataset.In the second stage，arobust consistency learning network with noise pseudo-label tolerant characteristics is constructed based on WideResNet backbone to implement ATR with high accuracy，in which a noise pseudo-label smoothing mechanism is designed as well as a piecewise noise pseudo-label tolerant loss function.Experiments are conducted on the moving and stationary target acquisition and recognition（MSTAR） SAR dataset. The experimental results demonstrate that the proposed method achieves an average recognition accuracy of 93.37% （204號 across lO-class targets with only five labeled training samples for each class，which significantly enhances recognition performance and generalization ability.

Keywords：synthetic aperture radar （SAR）；automatic target recognition （ATR）；semi-supervised； deep learning（DL）；pseudo-label

0 引言

成像雷達(dá)系統(tǒng)，廣泛部署在無人機、飛機、衛(wèi)星等移動平臺上。（剩余23882字）