改进型脉冲耦合神经网络高分辨率SAR图像分割

针对高分辨率SAR (合成孔径雷达)图像噪声强, 目标分割难度大的特点, 提出一种改进的脉冲耦合神经网络(pulse coupled neural network, PCNN)模型的SAR图像分割算法。首先根据SAR图像中相干斑噪声的特点, 采用复小波进行去噪。然后, 在传统PCNN模型的基础上, 对神经元的输入信号, 尤其是链接系数和 阈值的非线性衰减子因子进行了改进和简化, 同时对链接强度系数β进行理论上的近似推导, 并减少人工设置的参数。最后, 通过最佳阈值对其结果进行二值化处理得到感兴趣的目标图像。实验结果表明, 改进后的算法运行效率提高, 自适应性增强。与传统算法相比, 区域一致性提高0.013, 区域的对比度提高0.015, 效果优于传统的PCNN算法, 为高分辨率SAR图像分割提供了一种新策略。

High Resolution SAR Image Segmentation Using Improved PCNN

An improved PCNN (pulse coupled neural network) model of SAR (synthetic aperture radar) image segmentation was proposed, which aimed at the characteristics of strong noise and difficult segmentation for high resolution SAR image. At first, it used a complex wavelet to reduce noise according to SAR image speckle noise characteristics. Then, it improved and simplified the input signal, especially for the link coefficient and decay factor of threshold based on traditional PCNN model. It gave a theoretical approximate derivation and reduced the artificial setting of parameters. Finally, it adopted the appropriate threshold to quantify the segmentation result to get binary image of object. The experimental results show that proposed algorithm improves the operational efficiency and enhances adapting ability. Compared with traditional methods, the regional consistency is improved by 0.013, the contrast of the region is improved 0.015. The proposed PCNN model is superior to the traditional PCNN algorithm and can provide a new strategy for high resolution SAR image segmentation.