Paper Info
Title
An Effective Microscopic Detection Method For Automated Silicon-Substrate Ultra-Microtome (Asum)
Abstract
Three-dimensional (3D) representation of whole-brain cellular connectomics is the fundamental challenge for brain-inspired intelligence. And orderly automatic collection of brain sections on the silicon substrate is essential for the 3D imaging of cerebral ultrastructure. With the self-designed automated silicon-substrate ultra-microtome, serial brain sections can be orderly collected on the circular silicon substrates. In order to automate the collection process and further improve the efficiency of section collection, the form-invariant "Single Shot MultiBox-Detector" is proposed to detect the brain sections and baffles in the field of view of the microscope. And the "Cycle Generative Adversarial Networks" data augmentation method is proposed to alleviate the problem of fewer samples of the collected microscopic image dataset. The experimental results suggest that the proposed detection method could effectively detect the foreground objects in the microscopic images.
Year
DOI
Venue
2021
10.1007/s11063-019-10134-5
NEURAL PROCESSING LETTERS
Keywords
DocType
Volume
Microscopic object detection, Deep learning, Data augmentation, Serial sections
Journal
53
Issue
ISSN
Citations 
3
1370-4621
0
PageRank 
References 
Authors
0.34
0
2
Name
Order
Citations
PageRank
Long Cheng1149273.97
Weizhou Liu200.34