An important point-of-care diagnostic technology for COVID-19 is x-ray imaging of the lungs. Here we present a novel deep learning training method which combines both supervised and reinforcement learning methodologies which allows transfer learning in a convolutional neural network (CNN). The method integrated hill-climbing techniques and stochastic gradient descent with momentum to train the CNN architectures without overfitting on small datasets. The model was trained using the Kaggle COVID-19 Chest Radiography dataset. The dataset consists of 219 COVID-19 positive images, 1341 normal images, and 1345 viral pneumonia images. Since training of a CNN can be affected by bias and depends on the limitations of available computing power, the data set was reduced to 219 images for each class. From each of the classes, 150 random images were used for training the CNN algorithm and the model was tested with 69 independent images. Transfer training was done on three models, namely, VGG-19, DenseNet-201, and NASNet. The DenseNet-201 architecture performed the best in terms of accuracy achieving an accuracy of 96.1%. The VGG-19 and DenseNet-201 had sensitivity of 91.3 % while NASnet had a slightly higher sensitivity of 92.8%. This shows that we can have high confidence of the classification results achieved by these models. These results show that deep learning methodologies can be used for identifying COVID-19 patients quickly and accurately.
With the increasing use of deep learning methodologies in various biomedical applications, there is a need for a large number of labeled medical image datasets for training and validation purposes. However, the accumulation of labeled datasets is expensive and time consuming. Recently, generative adversarial networks (GAN) have been utilized to generate synthetic datasets. Currently, the accuracy of generative adversarial networks is calculated using a structural similarity index measure (SSIM). SSIM is not adequate for comparison of images as it underestimates the distortions near hard edges. In this paper, we compare the real DRIVE dataset to the synthetic FunSyn-Net using Fourier transform techniques and show that Fourier behavior is quite different in the two datasets, especially at high frequencies. It is observed that for real images, the amplitude of the Fourier components exponentially decreased with increasing frequency. For the synthesized images, the rate of decrease of the amplitude is much slower. If a linear function is fit to the high frequency components, the slope distributions for the two datasets are completely different with no offset. The average slope in the log scale for DRIVE dataset and FunSyn-Net were 0.0195, and 0.009 respectively. We also looked at auto correlations for the horizontal cut of the Fourier transform and again saw a statistically significant difference between the means for the two datasets. Finally, we also observed that Fourier transforms with real images have higher magnitude squared coherence as compared to the synthesized images. Fourier transform has shown great success for finding differences between real and synthesized images and can be used to improve the synthesized GAN models.
A machine learning algorithm combining reinforcement learning and supervised learning is demonstrated for training of near infrared spectroscopy data for non-destructive measurement of fruit quality. The model optimizes the combination of pretreatment methods, discriminant methods and calibration methods and also the parameters used in the methods to achieve highest prediction correlations. The model achieves better results than manual combinations of the previously demonstrated models.
A very low cost multispectral detector is developed and bench marked with full spectrometer measurements by measuring internal quality parameters of kiwis. The multispectator detector uses self-referenced reflectance to reduce measurement variations. It is demonstrated that even when using only twelve wavelengths, only a small loss of accuracy occurs with respect to a spectrometer in measurements of solid soluble content and dry matter. Further, using classification, similar accuracy is achieved in placing the fruits in bins based on their quality parameters. The measurements are rapid (less than 5 seconds), non-destructive and the system costs less than $50.
We designed and built a low-cost imaging spectrometer using an in-house grating and a webcam and demonstrated its applications for active learning in science with experiments ranging from understanding light spectra from various sources to detecting adulteration in edible oils. The experiments were designed and run in an elementary school in Waterloo, Ontario with young students from grade 4 to grade 8. The performance of the spectrometer is benchmarked to commercial spectrometers and showed excellent correlation for wavelengths between 450 nm to 650 nm. The spectral range can be improved by removing infra-red filters integrated in webcams.
Multispectral sensing is specifically designed to provide quantitative spectral information about various materials or scenes. Using spectral information, various properties of objects can be measured and analysed. Microscopy, the observing and imaging of objects at the micron- or nano-scale, is one application where multispectral sensing can be advantageous, as many fields of science and research that use microscopy would benefit from observing a specimen in multiple wavelengths. Multispectral microscopy is available, but often requires the operator of the device to switch filters which is a labor intensive process. Furthermore, the need for filter switching makes such systems particularly limiting in cases where the sample contains live species that are constantly moving or exhibit transient phenomena. Direct methods for capturing multispectral data of a live sample simultaneously can also be challenging for microscopy applications as it requires an elaborate optical systems design which uses beamsplitters and a number of detectors proportional to the number of bands sought after. Such devices can therefore be quite costly to build and difficult to maintain, particularly for microscopy. In this paper, we present the concept of virtual spectral demultiplexing imaging (VSDI) microscopy for low-cost in-situ multispectral microscopy of transient phenomena. In VSDI microscopy, the spectral response of a color detector in the microscope is characterized and virtual spectral demultiplexing is performed on the simultaneously-acquired broadband detector measurements based on the developed spectral characterization model to produce microscopic imagery at multiple wavelengths. The proposed VSDI microscope was used to observe colorful nanowire arrays at various wavelengths simultaneously to illustrate its efficacy.
Safe drinking water is essential for human health, yet over a billion people worldwide do not have access to safe drinking water. Due to the presence and accumulation of biological contaminants in natural waters (e.g., pathogens and neuro-, hepato-, and cytotoxins associated with algal blooms) remain a critical challenge in the provision of safe drinking water globally. It is not financially feasible and practical to monitor and quantify water quality frequently enough to identify the potential health risk due to contamination, especially in developing countries. We propose a low-cost, small-profile multispectral (MS) system based on Digital Holographic Microscopy (DHM) and investigate methods for rapidly capturing holographic data of natural water samples. We have developed a test-bed for an MSDHM instrument to produce and capture holographic data of the sample at different wavelengths in the visible and the near Infra-red spectral region, allowing for resolution improvement in the reconstructed images. Additionally, we have developed high-speed statistical signal processing and analysis techniques to facilitate rapid reconstruction and assessment of the MS holographic data being captured by the MSDHM instrument. The proposed system is used to examine cyanobacteria as well as Cryptosporidium parvum oocysts which remain important and difficult to treat microbiological contaminants that must be addressed for the provision of safe drinking water globally.
The evanescent field–based polymeric planar waveguide refractive index sensors having a high Q Fabry–Pérot (FP) cavity between identical dual Bragg gratings corrugated on the surface of the planar waveguide were developed. The FP Bragg gratings cavity was fabricated with a cavity size of 5 and 7 mm, respectively. The spectra of light reflected from fabricated Bragg gratings, which were butt joined, were measured and compared with different indices of surrounding media. It was obtained that the FP Bragg gratings cavity is more sensitive than the single Bragg grating. The sensor developed shows much promise in the application of biomedical diagnostics such as a biosensor and/or environmental monitoring systems.
This study demonstrates the high sensitivity of high Q polymeric planar waveguide refractive index sensors used on the
evanescent field. A Fabry-Perot Bragg gratings cavity was fabricated with a cavity size of 5 mm and 7 mm, respectively.
The spectra of light reflected from fabricated Bragg gratings, which were butt-joined, were measured and compared with
different indices of surrounding media. It was confirmed the FP Bragg gratings cavity is more sensitive than the single
Bragg grating. The sensor developed in this study shows much promise in the application of biomedical diagnostics such
as a bio-sensor and/or environmental monitoring systems.
In this paper, we present our recent measurement of second harmonic generation (SHG) from silicon nanowires which are vertically aligned. The SHG shows a great enhancement due to the increase of the surface area which breaks the symmetry of silicon lattice and increase the surface SHG. A high SHG is also obtained in counter polarization for both S and P polarization excitation. An enhancement of 80 times is also observed. This huge enhancement opens the door for novel applications including frequency mixing and frequency generation for various novel nonlinear application of silicon based devices.
We demonstrate an etched-core fiber Bragg grating sensor for detection of bio-chemical agents. The fiber Bragg grating
of the sensor is etched to a diameter of 7 μm. The transition between the etched and the unetched core consists of an
asymmetric taper resulting in excitation of multiple modes. The different excited modes respond differently to change in
refractive index, temperature and strain. This allows for measurements for changes in these three parameters in a single
measurement by simultaneous measurement of reflections in Bragg wavelengths for different modes. This parametric
discrimination is confirmed experimentally by measuring the refractive index of water as temperature is increased. The
sensor is then integrated in a micro-fluidic channel fabricated using Polydimethylsiloxane (PDMS) substrate and tested
by introducing different chemicals. The sensitivity of the sensor to refractive index change is 92 nm/riu close to the
refractive index of water. Assuming a wavelength resolution of 1 pm, index resolution of 1x10-5, a strain resolution of 1
microstrain, and a temperature resolution of 0.032 ºC is achieved by the sensor.
In incoherent-injected WDM-PONs, it is shown through simulations and experiments that Flat-band athermal AWGs
with channel spacing of 100GHz have improved performance in terms of BER than their Gaussian-shape counterpart,
due to the effect of signal filtering. At a BER of 10-9, the power penalty is 2 dB for Flat-band AWGs and more than 4
dB for Gaussian AWGs, which do not reach the required BER. In addition, the effect of AWG detuning in the WDM-PON
systems is investigated. It is shown that by detuning the two Flat-band AWGs by 40GHz from their center
wavelength, an additional 0.5dB power penalty is induced.
A simple and straightforward approach was devised to fabricate microspherical aggregates of alumina nanoparticles by self-assembly. Two different organosilanes were used as surfactant agents. Spherical aggregates were spin-coated on a silicon substrate. Spheres with diameters ranging from 100 nm to a few micrometers were thus assembled within an hour. The spatial distribution and the size distribution of the spheres are adjustable by changing the concentration of used organosilane. Effects of different organosilane agents, mixture molarity, and temperature on the size distribution of the spheres were also investigated.
All-optical header recognition using a tree-structure is reported for a three-bit address. Each bit of a three bit header is
read using an optical Sagnac AND gate and the outcome is used to control each level of the three level tree-structure
switch. Traffic at 10 Gb/s (payload) is directed through the switch and each possible address outcome is validated.
Reflective semiconductor optical amplifiers (RSOAs) are used as the 1 x 2 space switch at each node of the tree-structure
switch. A noise propagation analysis that considers mostly amplified spontaneous emission noise is presented. This
analysis takes into account the saturation power of the SOAs, their noise figure, the gain of each SOA and the coupled
optical power. It is concluded that large switches based on semiconductor optical amplifiers can be constructed using
cascaded SOAs.
In this paper we describe empirical models for predicting the performance of high power lasers,
semiconductor optical amplifiers, and superluminescent diodes. The utility of the models is verified by
comparing predicted results to actual performance of devices. Based on the model, three important
parameters are identified for improving the performance of high power devices. These parameters include
reducing the thermal resistance, reducing the series resistance, and reducing the vertical carrier leakage. A
method is described to measure the thermal resistance. We further describe experiments done to reduce the
series resistance of devices to achieve a value of less than 0.5 &OHgr; for a 1 mm long ridge device. Finally the
effect of carrier stopper layers is described to reduce vertical leakage of carriers.
The need for efficient detection of biochemical agents is becoming more compelling. High sensitivity chemical and biological sensors, based on etched core fiber Bragg gratings that detect change in the index of refraction of surrounding solutions, were developed to measure the index of refraction of different solutions. A sensitivity of 1394 nm/riu was achieved for a core diameter of 3.4 μm. Assuming an experimental wavelength resolution of 0.01 nm, we were able to detect an index change of 7.2⋅10-6. These chemical sensors, properly sensitized using common glutaraldehyde chemistry, can be effectively used as biosensors. A 20 base single stranded DNA of concentration of 0.7 μg/ml was successfully detected. Some ideas are also proposed for sensitivity improvement of the sensor, like using the higher order modes. It is shown that the higher order modes can also provide information about the cause of the fundamental shift, whether it is due to refractive index change or due to external effects like temperature and stress.
We report on half-Watt level single spatial mode superluminescent laser diode at 1335 nm. Output optical power in excess of 500 mW from a single facet of angle-striped waveguide was realized at 10°C of heatsink temperature with peak electro-optical efficiency of 28%. To our knowledge this is the highest optical power and electro-optic conversion efficiency in a SLED device reported so far in the literature. Further optimization could lead to revolutionary result: 1) the creation of a high power optical device (SLED) with electro-optical efficiencies approaching and/or exceeding that of Fabry-Perot lasers (counting both facet outputs) with absolute optical power levels compared to that of Fabry-Perot lasers, 2) Electro-optical efficiencies approaching internal quantum efficiencies which could well exceed the 70-80% range observed in present commercial semiconductor laser and light-emitting structures.
There is a need for integrating various active and passive devices on a single substrate to increase the functionality of optical modules [1-4]. One of the methods is to use regrowth for creating a low-loss passive waveguide butt-coupled to the active waveguide [1]. Besides being a complex technology, issues like low-loss coupling over multiple runs is still a challenge. A second technique used for integration is selective area growth [2]. In general, this technology does not allow for freedom in the design of the various layer thickness and bandgaps of the integrated waveguides. Quantum well interdiffusion [3] has also been used for integration by altering the bandgaps of the waveguides but also suffers from a lack of freedom in waveguide design and in the selection of the proper bandgaps.
Expanded mode alignment tolerant optical structures will play an important role in low-cost, large-scale packaging of optoelectronic devices. In this paper, we present two expanded mode structures for operation at 1.55 micrometers . Our devices use single epitaxial growth and conventional fabrication schemes. High butt-coupling efficiencies (> 40%) to a single mode fiber with relaxed alignment tolerances were achieved. The first of our devices uses adiabatic transformation over 500 micrometers . The second device uses resonant coupling over a much shorter region of 200 micrometers . The second scheme offers an interesting possibility for monolithic integration of active-passive components. We present the design and simulation results of such an integrated device.
We report on two techniques developed at the University of Maryland, College Park for fabricating expanded mode laser arrays. Both of these techniques use single epitaxial growth and conventional fabrication techniques. The first of these techniques is based on adiabatic mode transformation from a tightly confined active waveguide to a loosely confined large underlying passive waveguide over a mode transmission region 500 micrometers long. The devices butt couple to a standard single mode fiber with a coupling loss of 3.6 dB and reduced farfield divergence angles of 22 degree(s) and 9 degree(s) in the transverse and lateral directions respectively. The excess mode transformation loss is 1.3 dB. The second device is based on a novel resonant coupling scheme between two waveguides of different dimensions and refractive indices. The mode is transformed over a taper length of 200 micrometers with excess mode transformation loss of 0.6 dB. Butt coupling efficiencies of 41% (3.8 dB coupling loss) is achieved to a standard single mode fiber. The farfield divergence angles achieved are 24 degree(s) and 13 degree(s) in the transverse and lateral directions respectively.
Intrinsically large mode semiconductor lasers and mode transformers monolithically integrated with semiconductor lasers, are two promising approaches for making alignment tolerant structures that can be used for passive alignment to single mode optical fibers. This technique, in conjunction with a recently developed silicon waferboard integration scheme, will significantly simplify the assembly process and the packaging of transmitter laser arrays. The passive alignment technique consists of octagonal electrodeposited copper bosses to physically register the laser chip with percussion etched inverted pyramidal receptacles and v-grooves in a silicon substrate.
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.