Search

With the widespread use of touch-screen devices, it is more and more convenient for people to draw sketches on screen. This results in the demand for automatically understanding the sketches. Thus, the sketch recognition task becomes more significant than before. To accomplish this task, it is necessary to solve the critical issue of improving the distinction of the sketch features. To this end, we have made efforts in three aspects. First, a novel multi-scale residual block is designed. Compared with the conventional basic residual block, it can better perceive multi-scale information and reduce the number of parameters during training. Second, a hierarchical residual structure is built by stacking multi-scale residual blocks in a specific way. In contrast with the single-level residual structure, the learned features from this structure are more sufficient. Last but not least, the compact triplet-center loss is proposed specifically for the sketch recognition task. It can solve the problem that the triplet-center loss does not fully consider too large intra-class space and too small inter-class space in sketch field. By studying the above modules, a hierarchical residual network as a whole is proposed for sketch recognition and evaluated on Tu-Berlin benchmark thoroughly. The experimental results show that the proposed network outperforms most of baseline methods and it is excellent among non-sequential models at present.

Objective: Patellofemoral arthritis is a common type of knee osteoarthritis and a prime cause of anterior knee pain and disability. Most of the existing research on knee osteoarthritis focuses on tibial-femoral arthritis, while studies on patellofemoral arthritis are relatively rare. This study aims to observe changes in osteochondral and subchondral bone structure over time in the patella and femoral trochlea in an animal model of spontaneous patellofemoral arthritis. Methods: A total of 24 1-, 3- or 5-month-old healthy female Hartley guinea pigs were used for experiments. No intervention was applied, and the mechanical pain threshold was assessed prior to euthanasia. Bilateral knee joints were collected in the animals at the different ages, and the patellofemoral joints were taken to evaluate the bone microstructure of patellofemoral articular cartilage and subchondral bone by macroscopy, histopathology and micro-computed tomography (micro-CT). Results: There was a significant difference in the severity of femoral trochlea injury assessed by the Macro score between 5- and 1-month-old groups (P

Swine manure activated carbon (SMAC) with self-co-doped metallic compounds and heteroatoms was prepared by combining KOH activation with pyrolysis. The structure, electrochemical performance and adsorption of SMACs prepared with various activation temperatures, activation times, alkali-carbon ratios and pickling steps were investigated. A hierarchical pore structure existed in all the SMACs, which showed excellent and significantly different properties, including a specific capacitance of 278 F/g, and a high SSA of 2321 m2/g (SMACHF), which retained 100% of its capacitance after 10,000 cycles at 5 A/g and a Pb2+ adsorption capacity of 211.8 mg/g at room temperature (SMACNON). The SMACs contained heteroatom functional groups and Si, Ca and Mg compounds, which did not produce pseudocapacitors but did increase the Pb2+ adsorption capacity and electrochemical cycling stability. The Pb2+ adsorption capacities of SMACNON, SMACPHF and SMACHF corresponded to the effect of the functional group content rather than the effect of the SSA. These porous carbons were derived from inexpensive swine manure, combine suitable porosity with metallic elements and heteroatom doping, and have promising uses in multifunctional applications.

In order to solve the problem of fault diagnosis of switching power supply in medical equipment due to the lack of drawings, a fault diagnosis method based on feature fusion is proposed. In this paper, long short-term memory network and one-dimensional convolution neural network in deep learning are used to extract features of different dimensions and to adaptively fuse them into feature vectors. A fault intelligent diagnosis model based on feature fusion is established. At the same time, Adam algorithm is used to adaptively optimize the model to realize automatic recognition of fault state of switching power supply. Finally, taking the switch power supply of Mindray T8 monitor as an example, it is verified that the proposed method can effectively and accurately identify the common faults of the switch power supply without drawings, and the accuracy and accuracy of the model are better than the fault diagnosis method based on a single feature.

In the production process of the oil cooling pipe, the roundness error of the pipe is mainly measured manually, which has the problem of low detection efficiency and inability to detect in real-time. Based on machine vision, this paper proposes a method for roundness detection and sorting of oil cooling pipes. By establishing the kinematics model of the robot SR7CL, then using Adams and Matlab to simulate kinematics and dynamics to verify the correctness of the model and ensure the accuracy of grasping. Taking the nozzle of the oil cooling pipe as the research object, the characteristics of the nozzle are highlighted through adaptive threshold segmentation, and the contour information are extracted by the Canny algorithm. Finally, used the least square method to detected the roundness error, and the cooling pipes are determined according to the roundness error. This system not only improves the quality and efficiency of detection, but also solves the real-time problems.

In order to improve the performance of passive positioning and make up for the shortcomings of traditional technology, such as less information, slow positioning speed and low positioning accuracy, this paper proposes a pure angle target positioning method based on the cubature Kalman filter(CFK). Firstly, the posterior probability density function is calculated by the determined volume points, without the knowledge of the Jacobian matrix which is complex and hard to be solved. In addition, this method can effectively improve the accuracy of positioning. Simulation results show that, compared with the traditional method, the positioning accuracy of this method is higher.

Objective: This study aimed to perform experiments to investigate the change trend in brain magnetic induction phase shift (MIPS) during hemorrhagic shock of different degrees of severity and to find the correlation between brain MIPS value and commonly used physiological indicators for detecting shock so as to explore a noninvasive method suitable for prehospital real-time detection of cerebral blood perfusion in hemorrhagic shock. Approach: The self-developed MIPS detection system was used to monitor the brain MIPS value in the whole process of hemorrhagic shock models of rabbits with different degrees of severity (control, mild, moderate, and severe) of shock in real time. Meanwhile, common physiological parameters, including arterial blood lactate (ABL), mean arterial pressure (MAP), heart rate (HR),core body temperature (CBT), regional cerebral blood flow (rCBF), and electroencephalogram (EEG), were also evaluated. Main results: The findings suggested that the brain MIPS value showed a downward trend in the shock process, and the decline degree of the MIPS value positively correlated with the severity of shock. Moreover, it showed a good detection and resolution ability in time/process and severity (P < 0.05). The MIPS values significantly correlated with ABL (P < 0.01), CBT (P < 0.01), and EEG (P < 0.05) at all four shock levels; with MAP (P < 0.05) and rCBF (P < 0.05) in the control, moderate, and severe groups; and with HR (P < 0.01) only in the severe group. Significance: The results demonstrated that the brain MIPS value has the capability of detecting hemorrhagic shock. The MIPS technique is a noninvasive method suitable for prehospital real-time detection of cerebral blood perfusion in hemorrhagic shock.

High-performance porous carbon electrode materials derived from pine nut shells are prepared through KOH, H3PO4 and steam activation. The different morphologies and structures of the porous carbons are studied to investigate the mechanisms of pore tailoring and the interaction effects among the activation agent, pore structure and properties of the porous carbons. The porous carbons are all excellent electrode materials for supercapacitors with high capacitance characteristics due to their well-developed pore structures. The electrochemical performance is strongly related to pore structure and distribution, followed by specific surface area (SSA). The 3D network porous structure of porous carbon benefits ion storage and electrolyte internal diffusion, resulting in superior electrochemical properties such as an SSA of 1471 m2/g, an energy density of 23.45 Wh/kg and a capacitance of 241 F/g. Moreover, after 10,000 cycles, 99% of capacitance was retained at 5 A/g in 6 M KOH aqueous solution. A feasible approach for high-value utilization of low-cost biomass waste is presented.

How to avoid the self-occlusion of a moving object is a challenging problem. An approach for dynamically avoiding self-occlusion is proposed based on the depth image sequence of moving visual object. Firstly, two adjacent depth images of a moving object are acquired and each pixel’s 3D coordinates in two adjacent depth images are calculated by utilizing antiprojection transformation. On this basis, the best view model is constructed according to the self-occlusion information in the second depth image. Secondly, the Gaussian curvature feature matrix corresponding to each depth image is calculated by using the pixels’ 3D coordinates. Thirdly, based on the characteristic that the Gaussian curvature is the intrinsic invariant of a surface, the object motion estimation is implemented by matching two Gaussian curvature feature matrices and using the coordinates’ changes of the matched 3D points. Finally, combining the best view model and the motion estimation result, the optimization theory is adopted for planning the camera behavior to accomplish dynamic self-occlusion avoidance process. Experimental results demonstrate the proposed approach is feasible and effective.

This paper presents a novel strategy for developing wind power in large-scale (multi-GW) wind farms in China. It involves combining oversized wind farms, large-scale electrical storage and long-distance transmission lines to deliver ‘baseload wind power’ to distant electricity demand centers. Baseload wind power is typically more valuable to the electric utility than intermittent wind power, so that storage can be economically attractive even in instances where the cost per kWh is somewhat higher than without storage. The prospective costs for this approach to developing wind power are illustrated by modifying an oversized wind farm at Huitengxile, Inner Mongolia. The site has an average power density of 580 W/m2 at 50 m hub heights and is located 500 km north of Beijing. Using locally mass-produced wind turbines there are good prospects that wind power would be cost-competitive with coal power, on a lifecycle cost basis, while providing substantial net environmental benefits. Finally, the institutional challenges related to the prospect of large-scale wind energy development are addressed. Especially important are policies aimed at developing the capacity for mass production of as much of this technology in China as is feasible. Promising instruments for speeding up the introduction of this technology include: (i) international joint ventures between foreign vendors and developers and Chinese manufacturers; and (ii) wind resource development concessions.