Search

The application of cross-dataset training in object detection tasks is complicated because the inconsistency in the category range across datasets transforms fully supervised learning into semi-supervised learning. To address this problem, recent studies focus on the generation of high-quality missing annotations. In this study, we first specify that it is not enough to generate high-quality annotations using a single model, which looks only once for annotations. Through detailed experimental analyses, we further conclude that hard-label training is conducive for generating high-recall annotations, whereas soft-label training tends to obtain high-precision annotations. Inspired by the aspects mentioned above, we propose a dynamic supervisor framework that updates the annotations multiple times through multiple-updated submodels trained using hard and soft labels. In the final generated annotations, recall and precision improve significantly through the integration of hard-label training with soft-label training. Extensive experiments conducted on various dataset combination settings support our analyses and demonstrate the superior performance of the proposed dynamic supervisor.

Hygroscopic properties of aerosol particles are important for determining aerosol size distributions, and thus determining scattering and absorption coefficients at ambient atmospheric conditions. In this study, hygroscopic properties of aerosol particles at an urban site in Guangzhou, China, were measured using a Hygroscopic Tandem Differential Mobility Analyzer (H-TDMA) system during the winter and the summer. The results show that the urban aerosols were composed of more-hygroscopic, less-hygroscopic and non-hygroscopic particles. For less-hygroscopic particles of 40–200 nm in diameter, the hygroscopic parameter κ LH was around 0.15. For more-hygroscopic particles, the κ MH was from 0.290 to 0.339 with a particle size from 40 to 200 nm. For non-hygroscopic particles, the κ NH was about 0.015. It was found that the number fraction of less-hygroscopic particles (NF LH ) was correlated with the atmospheric oxidation which can be presented by OC/EC. This paper attributed the larger NF LH in winter to the higher value of OC/EC (3.0). Such conditions may lead to more formation of less-hygroscopic particles. Backward trajectories cluster analysis shows that there is a certain link between air mass origin and aerosol hygroscopicity, but it seems to be independent of the level of pollution. The difference of NF NH indicates that the mixing state of aerosol particles can also be affected by air mass origin. Diurnal variations in aerosol hygroscopic parameters in both seasons show that during daytime, aerosol particles tend to have a low degree of external mixing or quasi-internal mixing, resulting in a higher NF MH and a larger κ mean ; during nighttime and early mornings, they tend to be mixed externally, resulting in a lower NF MH and a smaller κ mean . This can be attributed to atmospheric aging effect and evolution of mixing layer height and implies that soot (non-hygroscopic) particles present to a large extent as internal mixtures by the time they leave the urban environment. Hence, only assuming a constant mixing state of soot particles, such as pure external or internal for the regional climate model and air quality model is still not realistic and may lead to uncertainties for the Pearl River Delta (PRD) region which is one of the three major economic regions in China. Comparing the diurnal variation of NF NH in both seasons, it seems that such a diurnal cycle was mainly related to the differences in evolution of mixing layer between two seasons. Such quantitative hygroscopic properties of sub-micro particles are essential in assessing their impact on weather–climate effect and atmospheric visibility.

Generally, multiple sensors are deployed to track a target synchronously in wireless sensor networks. However, asynchronous measurements exist intrinsically in multi-rate multi-sensor systems. Asynchronous measurements may also emerge in acoustic sensor networks, owing to the low propagation speed of acoustic signals. In order to handle the target tracking problem with asynchronous measurements, a nonlinear smoothing algorithm based on the fixed-point smoother and the square-root cubature Kalman filter is derived and applied in asynchronous wireless sensor networks for the first time. The estimation precision of the states increases along with the smoothing process, and a sensor can always obtain the optimal estimate of a state before its own next measurement by using the proposed algorithm. The numerical simulations demonstrate that, thanks to the smoothing effect of the fixed-point smoother, the proposed algorithm can obtain not only the remarkable position estimation results of the target, but also even better velocity estimation results. In addition, the proposed algorithm can obtain much more states' estimates than benchmark synchronous target-tracking algorithms, under the same condition of measurement count and communication cost.

High performance computation is critical for brain-machine interface (BMI) applications. Current BMI decoding algorithms are always implemented on personal computers (PC) which affect the performance of complex mapping models. In this paper, an FPGA implementation of Kalman filter (KF) algorithm is proposed as a new computational method. The neural ensemble activities are recorded from motor cortex of rats performing a lever-pressing task for water reward. Kalman filter, which is used for mapping neural activities to kinematic variables, is implemented both on PC (MATLAB-based) and FPGA. In FPGA architecture, the row/column-based method is adopted for the matrix operation instead of the traditional element-based method, parallel and pipelined structures are also used for efficient computation at the same time. The results show that the FPGA-based implementation runs 24.45 times faster than the PC-based counterpart while achieving the same accuracy. Such a hardware-based computational method provides a tool for high-performance computation, with profound implications for portable BMI application.