In recent years, to reduce the workload on operators in annotating hyperspectral images (HSI), researchers have started to employ few-shot learning (FSL) methods for HSI classification. During this process, it is typically necessary to combine spatial and spectral features of HSI into a single vector representation, a fusion that often compromises the detail of spatial features. To optimize the model’s performance in FSL using spectral and channel information, this study has designed an FSL network based on the reconstruction of spectral-spatial features (S-SRFSL). The network comprises two main stages: In the first stage, we introduce an innovative loss function to prevent channel aggregation phenomena during HSI channel selection; in the second stage, we utilize the HSI results obtained from the first stage to extract spatial, spectral, and joint spectral-spatial features. These three types of feature vectors are then combined into a feature matrix for reconstruction, facilitating classification tasks in a new feature space. The study selected four publicly available HSI datasets for model pre-training and two for testing. Experimental results demonstrate that S-SRFSL proposed in this study achieves superior performance on both test datasets.

Recently proposed decoupled training methods emerge as a dominant paradigm for long-tailed object detection. But they require an extra fine-tuning stage, and the dis-jointed optimization of representation and classifier might lead to suboptimal results. However, end-to-end training methods, like equalization loss (EQL), still perform worse than decoupled training methods. In this paper, we re-veal the main issue in long-tailed object detection is the imbalanced gradients between positives and negatives, and find that EQL does not solve it well. To address the problem of imbalanced gradients, we introduce a new version of equalization loss, called equalization loss v2 (EQL v2), a novel gradient guided reweighing mechanism that re-balances the training process for each category independently and equally. Extensive experiments are performed on the challenging LVIS benchmark. EQL v2 outperforms origin EQL by about 4 points overall AP with 14 ∼ 18 points improvements on the rare categories. More importantly, it also surpasses decoupled training methods. With-out further tuning for the Open Images dataset, EQL v2 improves EQL by 7.3 points AP, showing strong generalization ability. Codes have been released at https://github.com/tztztztztz/eqlv2

Object recognition techniques using convolutional neural networks (CNN) have achieved great success. However, state-of-the-art object detection methods still perform poorly on large vocabulary and long-tailed datasets, e.g. LVIS. In this work, we analyze this problem from a novel perspective: each positive sample of one category can be seen as a negative sample for other categories, making the tail categories receive more discouraging gradients. Based on it, we propose a simple but effective loss, named equalization loss, to tackle the problem of long-tailed rare categories by simply ignoring those gradients for rare categories. The equalization loss protects the learning of rare categories from being at a disadvantage during the network parameter updating. Thus the model is capable of learning better discriminative features for objects of rare classes. Without any bells and whistles, our method achieves AP gains of 4.1% and 4.8% for the rare and common categories on the challenging LVIS benchmark, compared to the Mask R-CNN baseline. With the utilization of the effective equalization loss, we finally won the 1st place in the LVIS Challenge 2019. Code has been made available at: https://github.com/tztztztztz/eql.detectron2.

The differential boost inverter has been widely used as a microinverter for the renewable distributed generations as it achieves an ac voltage with amplitude larger than the dc input. However, the unlimited voltage gain of the boost inverter is practically unachievable due to the existence of the parasitic resistance of the input inductor, which narrows the input voltage range and distorts the output voltage waveform as the turn-on duty ratio of the power switches approaches unity. To solve this problem, a novel coupled-inductor-based differential boost inverter is proposed, which uses a two-winding or three-winding coupled inductor to enhanced voltage gain. As a result, the input voltage range of the proposed inverter is significantly widened, facilitating grid-connected applications with a very low dc input voltage, compatible to the outputs of most PV panels. Moreover, the leakage inductance energy of the coupled inductor is recycled with voltages across the power switches clamped to the dc link capacitor by using a lossless diode. A prototype with 30 V dc input, 155 V ac output, nominal power 200 W, was built, which achieves a high voltage-boost, while keeping relatively low voltages stresses on the semiconductor devices.

LCL-filtered three-level inverters have been widely used in PV applications because of their high efficiency, high power density, and low cost. In practice, the inverter-side current feedback (ICF) control has dominated the PV market due to its low-cost property. However, the grid-side current becomes more sensitive to the grid distortion owing to its indirect regulation in an ICF controlled system. To address this issue, this paper proposes a new control structure to improve the power factor and eliminate the low-order harmonics of the grid-side current simultaneously. By combining the feedforward and proportional feedback of the capacitor current, this method can effectively eliminate the reactive power induced by the capacitor of an LCL-filter. In addition, the low-order harmonics can also be eliminated if the capacitor current is sent back to the resonant controller. Effectiveness of the proposed method has finally been verified with a 5kW LCL-filtered three-phase T-type PV inverter.

Single-phase split-source inverter is an attractive single-stage topology, as it can realize buck-boost function in a very compact way. However, this topology suffers from higher voltage stresses on dc-link than two-stage structure, especially under low ac gain conditions. This paper therefore analyzes the substantial reason of this problem. Then, based on the analyzing result, a general structure is proposed to solve the problem, and one practical implementation based on reverse-blocking IBGT is presented. Besides, a very simple SPWM modulation scheme is proposed together to improve its performance. Finally, compared with the existing SSI and its extensions, the proposed circuit can achieve higher ac gain with the same dc-link voltage. Its working principle and merits are explained in detail and also verified by experimental results.

The dynamic motion of fluids is essential in generating aesthetically appealing effects like the oriental ink painting and fluid art. Due to the prohibitively high cost required in volumetric fluid simulations, implementing an interactive volumetric painting system in immersive virtual environments (IVEs) is challenging. We propose a framework to generate immersive fluid dynamic environments by solving the Navier-Stokes equation on the viewing sphere. With this approach, we largely reduce the complexity without losing the effective resolution. We demonstrate our method on a real-time 360-degree painting system and verify the usability of our interface prototype with examples.

In wireless network, the resource limitations and process speed are the main constraints. They make many traditional background subtraction methods unable to be applied in wireless network. Vibe algorithm is a newly proposed background subtraction technique and its computational load is very low. That makes it available in the wireless network. However, Vibe is a pixel-level algorithm and it does not take the information of region into consideration. So it can be easily influenced by the background disturbance, noise and illumination change. In this paper a novel background subtraction algorithm which can be used in wireless network is proposed. To deal with these problems, the proposed algorithm employs graph-based image segmentation to segment every frame and construct the region-level information. With the region-level information, the proposed algorithm provides several strategies such as classification of regions, modification in neighboring regions and fast elimination of ghost region to improve the result of Vibe algorithm. Experiment results demonstrate that the proposed method can effectively extract the moving objects in various categories of scenes.