Search

One of the most important factors in training object recognition networks using convolutional neural networks (CNNs) is the provision of annotated data accompanying human judgment. Particularly, in object detection or semantic segmentation, the annotation process requires considerable human effort. In this paper, we propose a semi-supervised learning (SSL)-based training methodology for object detection, which makes use of automatic labeling of un-annotated data by applying a network previously trained from an annotated dataset. Because an inferred label by the trained network is dependent on the learned parameters, it is often meaningless for re-training the network. To transfer a valuable inferred label to the unlabeled data, we propose a re-alignment method based on co-occurrence matrix analysis that takes into account one-hot-vector encoding of the estimated label and the correlation between the objects in the image. We used an MS-COCO detection dataset to verify the performance of the proposed SSL method and deformable neural networks (D-ConvNets) as an object detector for basic training. The performance of the existing state-of-the-art detectors (DConvNets, YOLO v2, and single shot multi-box detector (SSD)) can be improved by the proposed SSL method without using the additional model parameter or modifying the network architecture., Submitted to International Conference on Image Processing (ICIP) 2018

Organic-inorganic halide perovskites for solar cell applications have generated tremendous research interest because of their unique photovoltaic properties and ease of fabrication. However, for this class of materials to be a true contender for commercial photovoltaics, durability and scaling up to large area need to be addressed without compromising performance. The first part of this thesis looks at fabrication methods for high efficiency devices based on HC(NH2)2PbI3 (FAPbI3), which are more thermally stable than CH3NH3PbI3 (MAPbI3). In particular the role of HI additive in the nucleation and crystallization of FAPbI3 is analysed resulting in the demonstration of high performance planar devices with negligible hysteresis. The durability of MAPbI3 and FAPbI3 cells with different designs are then compared by subjecting these planar cells to thermal stress to determine degradation activation energies. Using an Arrhenius model, degradation rates are then used for the prediction of cell lifetime. The method is rapid in identifying weak components in the cell that cause degradation and is effective in reporting relative durability between different perovskite materials and between different hole transport materials. The degradation of FAPbI3 at different relatively humidity levels is also studied. It is revealed that the formation of non-perovskite phases initiates at the grain boundaries. A post thermal treatment is developed to remove chemical residues at the grain boundaries, which effectively slows the degradation process. The last part of the thesis tackles the two challenges associated with large area perovskite solar cell fabrication. To tackle the issue of film non-uniformity, an anti-solvent spray process is developed which is more superior than the conventional anti-solvent drop process in terms of film uniformity and material quality for carrier transport. The best cell on 1.2 cm2 fabricated by this method achieved a certified power conversion efficiency of 18 %. To tackle the resistive loss as cell area increases, a metal grid is designed guided by modelling. Its implementation results in the demonstration of a large area 16 cm2 cell with a certified efficiency at 12.1 % which is the highest certified for a >10 cm2 perovskite device at the cell (not module) level.

Metal halide perovskite solar cells (PSCs) have undergone rapid progress. However, unstable performance caused by sensitivity to environmental moisture and high temperature is a major impediment to commercialization of PSCs. In the present work, a low-temperature, glass–glass encapsulation technique using high performance polyisobutylene (PIB) as the moisture barrier is investigated on planar glass/FTO/TiO2/FAPbI3/PTAA/gold perovskite solar cells. PIB was applied as either an edge seal or blanket layer. Electrical connections to the encapsulated PSCs were provided by either the FTO or Au layers. Results of a “calcium test” demonstrated that a PIB edge-seal effectively prevents moisture ingress. A shelf life test was performed and the PIB-sealed PSC was stable for at least 200 days. Damp heat and thermal cycling tests, in compliance with IEC61215:2016, were used to evaluate different encapsulation methods. Current–voltage measurements were performed regularly under simulated AM1.5G sunlight to monitor chan...

Elucidating the decay mechanisms of photoexcited charge carriers is key to improving the efficiency of solar cells based on organo-lead halide perovskites. Here we investigate the spectral dependence (via above-, inter- and sub-bandgap optical excitations) of direct and trap-mediated decay processes in CH3NH3PbI3 using time resolved microwave conductivity (TRMC). We find that the total end-of-pulse mobility is excitation wavelength dependent - the mobility is maximized (172 cm(2) V(-1) s(-1)) when charge carriers are excited by near bandgap light (780 nm) in the low charge carrier density regime (10(9) photons per cm(2)), and is lower for above- and sub-bandgap excitations. Direct recombination is found to occur on the 100-400 ns timescale across excitation wavelengths near and above the bandgap, whereas indirect recombination processes displayed distinct behaviour following above- and sub-bandgap excitations, suggesting the influence of different trap distributions on recombination dynamics.

The smart grid is a developing network of transmission lines, equipment, controls and new technologies working together to respond immediately to 21st Century demand for electricity, but the smart grid may also create much new vulnerability if not deployed with the appropriate security controls. In this paper, to evaluate the secure GOOSE message performance, we implemented the IEC 62351-6 MAC mechanism, applied it on F-IED using the IEC 61850 GOOSE stack and the HSM, built the security test environment, and obtained test results.

This paper provides deep understanding of the formation mechanism of perovskite film fabricated by sequential solution-based methods. It compares two sequential spin-coating methods for Cs0.15 (MA0.7 FA0.3 )0.85 PbI3 perovskite. First is the "static process," with a stoppage between the two spin-coating steps (1st PbI2 -CsI-dimethyl sulfoxide (DMSO)-dimethylformamide (DMF) and 2nd methylammonium iodide (MAI)-formamidinium iodide (FAI)-isopropyl alcohol). Second is the "dynamic process," where the 2nd precursor is dispensed while the substrate is still spinning from the 1st step. For the first time, such a dynamic process is used for Cs0.15 (MA0.7 FA0.3 )0.85 PbI3 perovskite. Characterizations reveal improved film formation with the dynamic process due to the "retainment" of DMSO-complex necessary for the intermediate phase which i) promotes intercalation between precursors and ii) slows down perovskite crystallization for full conversion. The comparison on as-deposited perovskite before annealing indicates a more ordered film using this dynamic process. This results in a thicker, more uniform film with higher degree of preferred crystal orientation and higher carrier lifetime after annealing. Therefore, dynamic-processed devices present better performance repeatability, achieving a higher average efficiency of 17.0% compared to static ones (15.0%). The new insights provided by this work are important for perovskite solar cells processed sequentially as the process has greater flexibility in resolving solvent incompatibility, allowing separate optimizations and allowing different deposition methods.

The influence of rapid-set accelerating admixtures on the setting behavior and early-age strength of a cement matrix was investigated to evaluate the appropriateness of the specification and test methods for shotcrete set accelerators. The results verified two different rapid setting behaviors according to the types of accelerator. The aluminate-base and the calcium aluminate cement-base accelerators facilitate hydration by formation of a calcium aluminate solid solution, whereas the alkali-free set accelerating agents present rapid setting time by the formation of ettringite. It was also found that the Vicat test was more desirable than the Gillmore test as the standard for setting time evaluation. Additionally, the cement mortar mixed with the aluminate-base and the calcium aluminate cement-base accelerators exhibited very fast development of early-age compressive strength. However, most of the set accelerators, except for alkali-free accelerators, failed to satisfy the specification because of greater than 40% compressive strength loss at 28 d.

Two-dimensional simulation studies are performed for modified laser wakefield acceleration. After one laser pulse, another identical laser pulse is sent to the plasma to amplify the wake wave resonantly. The simulation results show that the number of injected electrons is bigger than that of the single pulse case and the beam energy is higher as well. In addition, increase of the transverse amplitude is noticed in the wake wave after the second laser pulse. This shows that the transverse motion of the wake wave enhances the wave breaking for strong injection and acceleration of electron beams.

An extended 1D kinetic model, proposed by Lontano et al. [Phys. Plasmas 9 (2002) 2562], is used to investigate physical conditions of the relativistic soliton in electron–positron–ion plasmas. The existence domains of the soliton were specified in the plane of normalized frequency (ω)–temperature (λ) with varying relative density of the positrons, α m . The existence domain in ω–λ plane becomes larger when α m increases because of increase of the Langmuir frequency due to positrons. With temperatures under the rest mass energy of electron, m e c 2 , even a small amount of ions is important to support the soliton form.

Although set accelerating agents are used generally in New Austrian Tunneling Method, the standards for test methods and quality of set accelerating agents are not prescribed domestically. In this study, the proprieties of the various standards and the characteristics of set accelerating agents for shotcrete were evaluated. The alkali contents of set accelerating agents based on silicate, aluminate and cement were higher than those of alkali-free ones. From the result, it is thought that the quality control of aggregate should be enhanced and that the number of test cycle of alkali-aggregate reaction should be increased. The setting times of cement paste with set accelerating agents based on silicate and alkali-free ones were different largely with mixing methods. Compressive strength of mortar with set accelerating agents based on silicate, aluminate and cement at one day satisfied the specifications of Korea Concrete Institute. However, the strength ratio compared to control mix at 28 days showed as except for the alkali-free set accelerating agents. As a results of setting time and strength test, the establishment of domestic standards that can reflect the characteristics of materials and construction methods of tunnels and that can increase quality of set accelerating agents is required immediately.

The characteristics of pulse compression during the laser pulse amplification using two counter-propagating laser pulses have been investigated using a one-dimensional fluid model of stimulated Raman backscattering. The front and the rear pulse widths of the amplified laser pulse are examined in terms of the initial pump and seed amplitude, and the ratio of the electron plasma frequency to the laser frequency. The pulse widths have a common scaling behaviour in time over the considered parameter regime. The scaling exponents of the pulse widths during broadening and compression are all the same as 1.0. The pulse compression scaling shows a reciprocal behaviour to the peak energy growth.