In Chinese medicine, eye diagnosis is essential for diagnosis and treatment. However, most current image-processing techniques focus on tongue diagnosis, and most foreign studies on ocular diagnosis focus on segmenting fundus vascular images. Moreover, most of the foreign studies on scleral vessels are focused on identification rather than on TCM discernment. Scleral vessels can significantly characterize the pathological features of the human body’s five internal and six internal organs. Scleral vessels are essential for the objective study of TCM visual diagnosis. However, due to the small size and complex structure of scleral vessels, it is difficult to extract them by existing methods effectively. To achieve more accurate scleral blood vessel extraction, we introduce the residual connection structure and CA-Module attention mechanism in the U2Net1 network to avoid the incompatibility between high-level and low-level features and enhance the information extraction by input fusion and feature extraction of RSU blocks. The experimental results show that Miou achieves an accuracy of 83.3%. The F1-score reaches 91.7%, which is more effective than the existing SOTA fundus vascular segmentation network FR-UNet2 for the experiments. According to the experimental results, Res-U2Net can segment sclerar vessels accurately. In future experiments, we will improve the vessel feature extraction network to increase its accuracy and gradually achieve better results.
Image segmentation is a critical technology in many fields, such as image processing, pattern recognition, and artificial intelligence. It is also the first and critical step in computer vision technology. Tongue diagnosis combined with deep learning for segmentation and extracting pathological features is relatively mature, but deep learning combined with TCM visualization is sporadic. First, We used the U2Net network1 for segmentation extraction of the sclera in this study. Where the U2Net1 network1 (based on PyTorch) relies on the extensive use of data enhancements to use the available annotation samples more efficiently, and compared with the U-Net network, the U2Net network1 updates an RSU module, each RSU module is a small U-net network,merging multiple U-Net outputs to get the merged Mask target. Finally, we applied classical CNN networks to evaluate the segmentation effect, introducing different evaluation metrics such as Miou, Precision, and Recall. We used the publicly available dataset UBIVIS.V12 for our experiments, where our Miou was as high as 97.3%, and U2Net achieved better results among all the networks, which laid the foundation for our subsequent segmentation and extraction of blood filament features.
A method for fabricating a thin film Fabry–Pérot (F–P) cavity via dual pressure-assisted acrylate AB glue was studied. The method pressurizes both the acrylate AB glue to generate bubbles and the subsequent inflated film to make it thinner and more sensitive. The F–P cavity was fabricated by attaching the ultrathin film to a glass tube end face and a single-mode fiber in the glass tube to form a two-beam interference model. On the basis of two cycles and several repeated experiments for the humidity gradient, the linear and stable characteristics, and the humidity sensitivity of the F–P cavity with a thin film was demonstrated. The humidity sensitivity was 80.68 pm/% RH and 82.17 pm/% RH under two rising humidity experiments. Acting as the humidity sensing material for the first time, the acrylate AB glue film of the F–P cavity shows good stability and repeatability. In addition, the dual pressure-assisted method introduces a way to produce fiber-optic sensors with ultrathin film.
A new fiber optic negative pressure sensor based on pressure-sensitive diaphragm sealed Fabry-Perot cavity which
cascaded to the fiber tip is proposed in this paper. The experiment result shows a highly sensitivity of 101.9 pm/kPa
within the measurement range of 10.0-100.4 kPa, the response and recovery time are, respectively, 4 and 10 second over
the negative pressure variation of 100.4-15.1 kPa and 20.0-110.4 kPa. According to the experiment results, the proposed
sensor possesses the advantages of high sensitivity, fast response and recovery time.
The nonlinearity and dispersion characteristics of a suspended core fiber with different fluid filling methods and
slot-structure-embedded core configuration is simulated and analyzed here. Firstly, four suspended core bismuth fibers
were filled with high nonlinear fluid CS2 fluid by selective filling method. With the selective filling method, due to the
mode redistribution effect, the nonlinearity of optical fiber is significantly enhanced, and the dispersion characteristics of
optical fiber can be customized more smoothly. Furthermore, an elliptical, nonlinear fluid-filled nanoscale slot structure
is embedded into the suspended Bismuth fiber core for mode distribution adjustment and nonlinearity enhancement,
while the cladding air holes are partially filled with nonlinear fluid CS2 for mode field compression and nonlinearity
enhancement. Finally, the fiber parameters and the slot size are tailored properly for adjusting the fiber dispersion
characteristic. Simulation results show that an ultrahigh nonlinear coefficient that reaching up to 12023W-1km-1 and a
flattened dispersion with fluctuation less than 22ps/Km/nm at C-band are possible after the optimization.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.