基于改进MobileFaceNet的人脸识别方法北大核心

针对目前大多数人脸识别算法参数多、计算量大,难以部署到移动端和嵌入式设备中的问题,提出了一种基于改进MobileFaceNet的人脸识别方法。通过对MobileFaceNet模型结构的调整,将bottleneck模块优化为sandglass模块,改良深度卷积和逐点卷积的相对位置,适当增大sandglass模块的输出通道数,从而减少特征压缩时的信息丢失,增强人脸空间特征的提取。实验结果表明:改进后的方法在LFW测试数据集上准确率达99.15%,模型大小和计算量分别仅为原算法的61%和45%,验证了所提方法的有效性。     

Soluble polyimide as liquid crystal perpendicular alignment layer

Journal of Materials Science - Three polyimides (PIs) were prepared from a diamine containing biphenyl ester group as a side chain and a corresponding dianhydride chosen from...     

Destination-aware metric based social routing for mobile opportunistic networks

Wireless Networks - Although centrality is widely used to differentiate the importance of nodes for social-aware routing in mobile opportunistic networks (MONs), it is destination-agnostic since...     

Dielectric performance and conductive mechanism of KNN-based ceramics prepared via sol–gel core–shell process

Journal of Materials Science: Materials in Electronics - 0.95(Li0.02Na0.50K0.48)(Nb0.95Sb0.05)O3–0.05AgTaO3@BaZrO3 (LNKNbSAT@BZ) lead-free ceramics were prepared via a sol–gel...     

Co-precipitation synthesis of ZnAl2O4: Cr3+ phosphor for better light penetration in pc-LED

Cr doped ZnAl₂O₄ spinel samples were prepared by the traditional solid state reaction and co-precipitation synthetic route, and the results suggest that the co-precipitation method has some superiority in contrast to the solid state reaction method. XRD, FT-IR, and XPS spectra confirmed that the well-crystallized spinel cubic phase of ZnAl₂O₄: Cr³⁺ samples were successfully formed. The morphology of the samples was investigated by FE-SEM and FE-TEM, and the results show that the samples by the co-precipitation route can generate a smaller size of particles compared to the solid state reaction. Photoluminescence excitation spectra monitored at 686 nm are comprised of two broad excitation bands near 530 nm and 395 nm, and the emission spectra show emissions ranging from 640 to 780 nm, due to the ²E?→?⁴A₂ spin-forbidden transition of Cr³⁺ ions in spinel lattices. The optimized concentration monitored at 686 nm is 1%, while at 693 nm is 3.5%. Compared with the samples by solid state reaction method, the samples by co-precipitation method show preferable luminescent properties, such as the higher photoluminescence intensity and higher quantum efficiency. Several phosphor-converted LEDs were to investigate the applicability of the prepared samples. The results confirm that the phosphor has potential applications in plant growth and supplementing the red region in white-LEDs and the phosphors prepared by co-precipitation are more suitable to be used in phosphor-converted LED devices due to their preferable luminescent properties.

     

Characteristic tryptic peptides and gelling properties of porcine skin gelatin affected by thermal action

Thermal action in extraction process had effects on characteristic tryptic peptides identification and gelling properties of porcine gelatin. SDS-PAGE, HPLC-LTQ/Orbitrap high-resolution mass spectrometry, texture analyser and rheometer were used to evaluate collagen depolymerisation degree, characteristic tryptic peptides and gelling properties of gelatins prepared in various thermal actions. Results showed that with increasing temperature and time, depolymerisation degree enlarged, while gel strength, gelling and melting temperature decreased. Mass spectra showed that 47 and 49 common characteristic tryptic peptides were identified in gelatins extracted at 50 °C and 100 °C with various times, respectively. Moreover, 34 common characteristic tryptic peptides were identified in all gelatin samples. Further comparison between this work and our previous investigations yielded 20 common characteristic tryptic peptides, which stably exist in various thermal actions. These common characteristic tryptic peptides may be very helpful for the accurate authentication of porcine gelatin.     

Co?N graphene encapsulated cobalt catalyst for H2O2 decomposition under acidic conditions

Hydrogen peroxide (H₂O₂) has been listed as one of the 100 most important chemicals in the world. However, huge amount of residual H₂O₂ is hard to timely decomposed into O₂ and H₂O under acidic condition, easily resulting in explosion hazard. Here, we reported a core–shell structure catalyst, that is graphene with Co N structure encapsulated Co nanoparticles. Co N graphene shell serves as the active site for the H₂O₂ decomposition, and Co core further enhance this decomposition. Benefiting from it, the H₂O₂ decomposition were close to 100% after 6 cycles without pH adjustment, which increased 6 orders of magnitude compared with no catalyst. At the same time, the O₂ generation reached 99.67% in 2 h with little metal leaching, and ·OH has been greatly inhibited to only 0.08%. This work can cleanly remove H₂O₂ with little deep oxidation and protect the process of H₂O₂ utilization to achieve a safer world.