独立源信号的卷积混合比线性混合更接近真实情况。利用非参数概率密度估计方法———Parzen窗函数估计法 ,提出了一种具有良好适应性的盲解卷积改进算法。该算法可以在无需知道信号分布形式的情况下 ,较准确地估计出密度函数值 ,且比传统的最大算法中采用固定的概率密度函数估计更接近信号点的真实概率密度。同时 ,此算法还具有无论对规则、单峰分布还是不规则、多峰分布都可以取得较好的估计的优点。因此 ,在理论上 ,改进算法可以获得比传统算法更优越的分离性能且能广泛地应用于具有各种分布的信号。实验结果证实 ,这一算法能有效地从各种分布的信号包括真实语音、图像等构成的卷积混合信号中恢复出原始信号。与最大熵算法相比 ,改进算法具有更好的分离性和更广泛的适用性。 The convolutional mixing ratio of independent source signals is closer to the real situation than linear mixing. Using nonparametric probability density estimation method - Parzen window function estimation method, an improved blind deconvolution algorithm with good adaptability is proposed. The algorithm estimates the density function more accurately without knowing the signal distribution, and uses a fixed probability density function to estimate the true probability density closer to the signal point than the traditional maximum algorithm. At the same time, this algorithm also has the advantage of obtaining a better estimation of the multi-peak distribution regardless of the rule, unimodal distribution or irregularity. Therefore, in theory, the improved algorithm can obtain more superior separation performance than traditional algorithms and can be widely applied to signals with various distributions. The experimental results show that this algorithm can effectively recover the original signal from all kinds of convolutional mixed signals composed of real voice, image and so on. Compared with the maximum entropy algorithm, the improved algorithm has better separability and wider applicability.