基于小波变换和深层稀疏编码的SAR目标识别

针对SAR图像预处理算法自适应能力差、带标签图像不足、目标特征提取困难等问题,提出了一种基于小波变换和深层稀疏编码的SAR图像目标自动识别算法。首先利用灰度值和尺度缩放获得大量的无标签SAR目标,并采用离散小波变换对图像进行高效的降维,再结合深层稀疏编码提取目标的深层抽象特征并完成识别任务。采用MSTAR数据库中3类军事目标进行算法仿真与验证。实验结果表明,在没有预处理的情况下,该算法能够有效地完成多目标SAR图像分类,且具有较高的识别率和鲁棒性。

SAR TARGET RECOGNITION USING WAVELET TRANSFORM AND DEEP SPARSE AUTOENCODERS

To overcome the low adaptability of the preprocessing algorithm, lack of labelled images and difficulty of target feature extraction, a novel approach to target recognition of SAR images which combines deep sparse autoencoders(DSA) and discrete wavelet transform(DWT) was presented in this paper. The gray value and scale variation was used for obtaining large amount of unlabeled SAR targets. The DWT was applied for dimensionality reduction of SAR images. Moreover, through the formation of deep sparse autoencoders, deep abstract feature was learned from the SAR targets. Experiments were implemented with three military targets in MSTAR database. Experimental results based on the MSTAR database demonstrate the proposed algorithm can accomplish the multiple-targets classification effectively even without preprocessing, and has a higher recognition rate and robustness.