ABSTRACT
As a popular research subject in the field of computer vision, knowledge distillation (KD) is widely used in semantic segmentation (SS). However, based on the learning paradigm of the teacher-student model, the poor quality of teacher network feature knowledge still hinders the development of KD technology. In this paper, we investigate the output features of the teacher-student network and propose a feature condensation-based KD network (FCKDNet), which reduces pseudo-knowledge transfer in the teacher-student network. First, combined with the pixel information entropy calculation rule, we design a feature condensation method to separate the foreground feature knowledge from the background noise of the teacher network outputs. Then, the obtained feature condensation matrix is applied to the original outputs of the teacher and student networks to improve the feature representation capability. In addition, after performing feature condensation on the teacher network, we propose a soft enhancement method of features based on spatial and channel dimensions to improve the dependency of pixels in the feature maps. Finally, we divide the outputs of the teacher network into spatial condensation features and channel condensation features and perform distillation loss calculation with the student network separately to assist the student network to converge faster. Extensive experiments on the public datasets Pascal VOC and Cityscapes demonstrate that our proposed method improves the baseline by 3.16% and 2.98% in terms of mAcc, and 2.03% and 2.30% in terms of mIoU, respectively, and has better segmentation performance and robustness than the mainstream methods.
ABSTRACT
Multiferroics are being studied increasingly in applications of photovoltaic devices for the carrier separation driven by polarization and magnetization. In this work, textured black silicon photovoltaic devices are fabricated with Bi6Fe1.6Co0.2Ni0.2Ti3O18/Bi2FeCrO6 (BFCNT/BFCO) multiferroic heterojunction as an absorber and graphene as an anode. The structural and optical analyses showed that the bandgap of Aurivillius-typed BFCNT and double perovskite BFCO are 1.62 ± 0.04 eV and 1.74 ± 0.04 eV respectively, meeting the requirements for the active layer in solar cells. Under the simulated AM 1.5 G illumination, the black silicon photovoltaic devices delivered a photoconversion efficiency (η) of 3.9% with open-circuit voltage (Voc), short-circuit current density (Jsc), and fill factor (FF) of 0.75 V, 10.8 mA cm-2, and 48.3%, respectively. Analyses of modulation of an applied electric and magnetic field on the photovoltaic properties revealed that both polarization and magnetization of multiferroics play an important role in tuning the built-in electric field and the transport mechanisms of charge carriers, thus providing a new idea for the design of future high-performance multiferroic oxide photovoltaic devices.
ABSTRACT
Multiferroic devices have attracted renewed attention in applications of photovoltaic devices for their efficient carrier separation driven by internal polarization, magnetization, and above-bandgap generated photovoltages. In this work, Zn2SnO4-based multiferroic Bi6Fe1.6Co0.2Ni0.2Ti3O18/Bi2FeCrO6 (BFCNT/BFCO) heterojunction photoelectrodes were fabricated. Structural and optical analyses showed that the bandgap of the spinel Zn2SnO4 is â¼3.1 eV while those of Aurivillius-type BFCNT and double-perovskite BFCO are 1.62 and 1.74 eV, respectively. Under the simulated AM 1.5G illumination, the as-prepared photoelectrodes delivered a photoconversion efficiency (η) of 3.40% with a short-circuit current density (Jsc), open-circuit voltage (Voc), and fill factor (FF) of 10.3 mA·cm-2, 0.66 V, and 50.4%, respectively. Analyses of adjustment of an applied electric and magnetic field on photovoltaic properties indicated that both magnetization and polarization of multiferroics can effectively tune the built-in electric field and the transport of charge carriers, providing a new idea for the design of future high-performance multiferroic oxide photovoltaic devices.
ABSTRACT
Bilayer magnetoelectric (ME) nanofilms composed of Bi0.9Nd0.1FeO3 (BNF) and Ni0.55Zn0.45Fe2O4 (NZF) were fabricated on the Pt(111)/Ti/SiO2/Si(100) substrates via sol-gel and a subsequent rapid thermal process with different growth sequences of BNF and NZF forming the following layered structures: BNF/NZF and NZF/BNF. The phase composition, microstructure, and ferroelectric, dielectric, ferromagnetic, and ME coupling properties of the composites were investigated at room temperature. Structural characterization by X-ray diffraction and scanning electron microscopy showed that there are no other impurity phases but BNF and NZF, and the nucleation barrier caused that it is easier for NZF and BNF to grow on each other rather than on the surface of Pt/Ti/SiO2/Si. The tests of the physical properties indicated that such heterostructures present both good ferroelectric, ferromagnetic, and dielectric properties and the in-plane ME coupling coefficient α E at room temperature but some discrepancies also exist, which can be attributed to an interfacial effect, in other words, the deposition sequences of the constituent phases have a great influence on the properties of bilayer films.