高级检索+

基于目标检测及边缘支持的鱼类图像分割方法

Fish Image Segmentation Method Based on Object Detection and Edge Support

  • 摘要: 对图像中的鱼类目标进行分割是提取鱼类生物学信息的关键步骤。针对现有方法对养殖条件下的鱼类图像分割精度较低的问题,提出了基于目标检测及边缘支持的鱼类图像分割方法。首先,设计了基于目标检测的完整轮廓提取方法,将具有完整轮廓的鱼类目标从图像中提取出来作为分割阶段的输入,使得整幅图像的分割问题转化为局部区域内的分割问题;然后,搭建Canny边缘支持的深度学习分割网络,对区域内的鱼类实现较高精度图像分割。实验结果表明,本文方法在以VGG-16、ResNet-50和ResNet-101作为主干网络的模型上的分割精度为81.75%、83.73%和85.66%。其中,以ResNet-101作为主干网络的模型与Mask R-CNN、U-Net、DeepLabv3相比,分割精度分别高14.24、11.36、9.45个百分点。本文方法可以为鱼类生物学信息的自动提取提供技术参考。

     

    Abstract: Segmenting fish objects from images is a key step in extracting fish biological information. In view of the low accuracy of current methods in fuzzy underwater fish image segmentation, a fish image segmentation method based on object detection and edge support was proposed. Firstly, the fish objects were cut out from the image by using the method of object detection, and the whole image segmentation was transformed into region segmentation. Then, the edge support method was used to segment the fish in the region, so as to further improve the segmentation accuracy of the model. The experimental results showed that the segmentation accuracy of the method was 81.75%, 83.73% and 85.66%, respectively by the models with VGG-16, ResNet-50 and ResNet-101 as the backbone network. The segmentation accuracy of the model with ResNet-101 as the backbone network was 14.24 percentage points, 11.36 percentage points and 9.45 percentage points higher than that of Mask R-CNN, U-Net and DeepLabv3 models, respectively. The method can be applied to the automatic extraction of fish biological information.

     

/

返回文章
返回