基于改进U-Net的活性污泥显微图像分割方法

Activated Sludge Microscopic Image Segmentation Method Based on Improved U-Net

针对活性污泥相差显微图像存在伪影且现有图像分割方法对丝状菌的分割精度不高等问题,提出一种以U-Net网络为基础,结合残差网络、通道注意力机制、空洞空间金字塔模块的活性污泥显微图像分割模型。使用带有通道注意力机制的ResNet网络作为编码器,通道注意力机制显式地建立了特征通道间的依赖关系,分析了残差网络强化模型的特征提取能力。通过在编码器的最后加入空洞空间金字塔池化,可在不增加参数量的同时获得丝状菌与絮体的多尺度信息。通过在解码器中使用跳跃连接来补充特征信息,强化网络的重建能力。实验结果表明,与U-Net、DeepLabV3+等算法相比,所提模型具有更好的分割性能与效果。

Aiming at the problems of artifact existing in activated sludge phase contrast microscopic images and low segmentation accuracy of existing image segmentation methods for filamentous bacteria,a segmentation model of an activated sludge microscopic image based on the U-Net network,residual network,channel attention mechanism,and atrous spatial pyramid module is proposed.The ResNet network based on channel attention mechanism is used as an encoder.Channel attention mechanism explicitly establishes the dependence among feature channels,and analyzes the feature extraction ability of the residual network reinforcement model.At the end of the encoder,the atrous spatial pyramid pooling is added,which can obtain the multi-scale information of filaments and flocs without increasing the parameters.In order to enhance the ability of network reconstruction,the feature information is supplemented by using jump connection in the decoder.Experimental results show that the proposed model has better segmentation performance and effect than U-Net and DeepLabV3+.