基于DeepLabV3+的污泥显微图像分块分割方法

SLUDGE MICROSCOPIC IMAGE CHUNKING BASED ON DEEPLABV3+ METHOD

  • 摘要: 针对传统图像方法处理活性污泥相差显微图像存在过分割、欠分割,甚至分割失败问题,基于改进的DeepLabV3+网络进行分块分割以提高丝状菌分割效果。该方法以一定重叠率将高分辨相差显微图像切分成多块区域并进行分割,再将分割图像拼接恢复到原始分辨率。所提方法在某城市污水处理厂活性污泥显微图像数据上进行验证。实验结果表明,轻量化分块分割方法相对于未分块的DeepLabV3+、U-Net、SegNet模型在精确率、召回率、像素准确率和IoU性能指标上有一定程度的提升,模型大小显著降低。

     

    Abstract: Activated sludge phase contrast microscopic image segmentation based on traditional image methods often suffers from over-segmentation, under-segmentation, and even segmentation failure. An improved DeepLabV3+ network is used for block segmentation to improve the segmentation effect of filamentous bacteria. This method divided a high-resolution phase-contrast microscopy image into multiple small areas with a certain overlap rate. The segmented images were stitched back to the original resolution. The proposed method was verified on the microscopic image data of activated sludge from a municipal sewage treatment plant. The experimental results show that the lightweight segmentation strategy model has a certain degree of improvement in segmentation accuracy, recall, pixel accuracy and IoU performance indicators compared with the undivided DeepLabV3+, U-Net, and SegNet models, and the model size is significantly reduced.

     

/

返回文章
返回