Research On C.elegans Image Segmentation Algorithm Based On Improved Mask R-CNN

Caenorhabditis elegans has simple but complete organs,and most of the physiological changes are usually estimated by visual observation and lack of quantification.Computer vision technology can realize automatic quantitative assessment,which helps researchers to quantitatively calculate and analyze various physiological changes of nematodes.This article mainly studies the segmentation of Caenorhabditis elegans microscopic images,which lays the foundation for the life expectancy,three-dimensional reconstruction and dynamic tracking of nematodes.Caenorhabditis elegans microscopic images are noisy,nematode edge pixels are similar to the surrounding environment,and the nematode’s posture has multiple interference factors such as flagella and attachments need to be separated.Traditional algorithms cannot meet the high-precision and robust segmentation requirements.In this paper,a new segmentation algorithm based on Mask Region-Convolutional Neural Networks is proposed.This algorithm performs automatic segmentation by learning nematode morphological features.Because the original network structure cannot learn context information and visual attributes well,the algorithm combines high-level semantic features with low-level edge features to achieve multi-level feature fusion,which can make full use of the context information in the network and add it to the mask branch The fully connected route assists FCN to generate higher quality segmentation masks.Because the nematodes in the microscopic images are mostly translucent,the model is not easy to distinguish them from the background.The algorithm also improves the loss function and the candidate box screening algorithm,which effectively reduces the false detection of multiple nematodes.Experimental results show that,compared with the original Mask R-CNN,the average accuracy rate(AP₅₀)of this method is improved by 4.3 percentage points,and the average merge ratio（mIOU）is improved by 4 percentage points.The algorithm results show that the deep learning segmentation method proposed in this paper can effectively improve the segmentation accuracy,and can segment nematodes more accurately under the microscopic image.