Abstract: In functional optical imaging，the very low SNR makes the sample covariance matrix have singularity，which brings the whitening the operation of the ICA denoising method to reduce the dimensionality，so the ICA denoising method will lost the signal detection ability.To resolve the problem，with the difference between the observed signals and the noise reconstructed by the initial ICA denoising method，a difference ICA（DICA） denoisng method is presented in this paper.Use the traditional filtering denoising method and the DICA denoisng method to deal with the simulated optical imaging datum in very low SNR，the results show that compared to the filtering denoising method，the DICA denoising method not only improves the denoising performance，but also can keep the output SNR while the input SNR changing.

Key words: function optical imaging, Independent Component Analysis（ICA）, filtering

摘要： 在光学功能成像中，极低信噪比会使得样本协方差矩阵具有奇异性，因此导致Emir等人提出的ICA去噪方法在白化预处理过程会出现降维现象，最终使得该方法无法检测出信号。为解决一问题，利用原ICA去噪方法得到的噪声信号与观测信号之间的差异特性，提出了一种比差ICA去噪方法，该方法在信噪比-40 dB情况下能成功检测出信号。利用仿真得到的光学功能成像数据，对比分析了比差ICA去噪方法与传统滤波去噪方法在极低信噪比下的检测性能，结果表明比差ICA去噪方法不仅检测性能明显优于滤波去噪方法，且输出信噪比基本不随输入信噪比的下降而下降。

关键词: 脑光学功能成像, 独立成分分析, 滤波