We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Motion Recovery and Segmentation of Biomedical Microscope Cell Images.
- Authors
Zijiang Zhu; Jianjun Li; Junshan Li; Junhua Wang; Yi Hu
- Abstract
In order to study the motion recovery and segmentation of biomedical microscopic cell images, firstly, the accuracy of motion estimation of the model was observed by selecting the conscious mice with limited brain. The speed was embedded in the HMM (Hidden Markov Model) to construct the motion recovery model of microscopic image and simulate it. Then, a high-resolution cell image database was set up for the images observed, and then a cell image segmentation model was constructed by using convolutional neural network, and it was simulated and evaluated. The results show that in the quantitative analysis of motion recovery by HMM and SEHMM (Hidden Markov Model with Speed Embedding), SEHMM can reduce the number of inconsistencies to a greater extent and its accuracy is higher than HMM. But when the speed is closer to zero from k-1 to k, it is basically a static stage, and then there is no significant difference between SEHMM and HMM. In the analysis of Unet network test, it is found that both the segmentation accuracy curve and the error curve can achieve high accuracy and small error when epoch is small. Whereas, with the increase of epoch, the accuracy and error will not change significantly, and it is likely that the network will be over-fitted and the computational burden of the software will increase. In the test and analysis of network model with different convolution layers, it is found that when the number of convolution layers is set to 4, the accuracy of image segmentation is the highest and the error is the lowest, and the effect is the most obvious. Therefore, through the research, it is found that the SEHMM model can be built by embedding speed into HMM to achieve better results. In image segmentation, high accuracy and low error can be achieved by reasonably arranging the number of convolution layers in the convolution neural network model. Although there are some problems in the research process, it still provides experimental basis for solving the problems of cell segmentation in later biomedical microscopy, and has important research significance.
- Publication
Acta Microscopica, 2020, Vol 29, Issue 1, p498
- ISSN
0798-4545
- Publication type
Academic Journal