基于改进DenseNet的电阻层析成像图像重建算法

Electrical resistance tomography imaging reconstruction algorithm based on improved DenseNet

针对电阻层析成像(ERT)的逆问题存在严重的病态性、非线性和欠定性,导致经典算法的重建图像通常精度偏低的问题,提出一种基于改进DenseNet网络优化的电阻层析成像重建算法。首先,采用Landweber算法迭代值作为图像重建初始解;其次,构建了融合CBAM注意力机制的多尺度卷积模块以获取不同尺度特征,从而加强对关键特征的提取;使用DenseNet作为图像重建的主干网络,引入Swish作为网络的激活函数并融合dropout算法提高网络的泛化能力;最后,使用余弦退火算法优化学习率,避免模型训练陷入局部最优。此外,对改进DenseNet网络进行了抗噪性实验和静态实验。实验结果表明,采用改进算法进行ERT图像重建,相对误差和相关系数均得到提升。该算法不仅具有较高的重建精度和良好的可视化效果,还表现出对抗噪声干扰的特性。

In view of the severe ill-posedness,nonlinearity and under-determinedness in the inverse problem of electrical resistance tomography(ERT),which often leads to low accuracy in image reconstruction of the classical algorithms,an ERT imaging reconstruction algorithm based on improved DenseNet is proposed.The Landweber algorithm is employed to generate the iterative value,which is taken as the initial solution for image reconstruction.A multiscale convolutional module incorporating the convolutional block attention module(CBAM)is constructed to capture features of different scales,thereby enhancing the extraction of key features.The DenseNet is utilized as the backbone network for image reconstruction,integrating Swish as the activation function of the network and incorporating the dropout algorithm to enhance the generalization capability of the network.The cosine annealing algorithm is used to optimize the learning rate of the model,preventing the model from getting stuck in local optima during training.Additionally,noise resistance experiments and static experiments are conducted to evaluate the performance of the improved DenseNet.The experimental results demonstrate that both the relative error and correlation coefficient are improved after employing the proposed algorithm for ERT image reconstruction.The algorithm not only achieves higher reconstruction accuracy and superior visualization,but also exhibits robustness against noise interference.