Towards Stabilized Few-Shot Object Detection with Less Forgetting via Sample Normalization.

Few-shot object detection is a challenging task aimed at recognizing novel classes and localizing with limited labeled data. Although substantial achievements have been obtained, existing methods mostly struggle with forgetting and lack stability across various few-shot training samples. In this paper, we reveal two gaps affecting meta-knowledge transfer, leading to unstable performance and forgetting in meta-learning-based frameworks. To this end, we propose sample normalization, a simple yet effective method that enhances performance stability and decreases forgetting. Additionally, we apply Z-score normalization to mitigate the hubness problem in high-dimensional feature space. Experimental results on the PASCAL VOC data set demonstrate that our approach outperforms existing methods in both accuracy and stability, achieving up to +4.4 mAP@0.5 and +5.3 mAR in a single run, with +4.8 mAP@0.5 and +5.1 mAR over 10 random experiments on average. Furthermore, our method alleviates the drop in performance of base classes. The code will be released to facilitate future research.

Diffusion-Ordered Spectroscopy (DOSY) as a Powerful NMR Method to Investigate Molecular Mobilities and Intermolecular Interactions in Complex Oil Mixtures.

Studying the characteristics and molecular mechanisms of liquid self-diffusion coefficient and viscosity changes is of great significance for, e. g., chemical and petroleum processing. As examples of highly complex liquid,an asphaltene-free high-acid and high-viscosity crude oil and its extracted fractions were studied by comparing their 1 H DOSY diffusion maps. The crude oil exhibited a polydisperse diffusion distribution, including multiple diffusion portions with diffusion coefficients much smaller than that of any single fraction in independent diffusion. The main mechanism that leads to the decreases in the diffusion coefficients of crude oil is attributed to diffusion resistance enhanced by Dynamical Molecular-Interaction Networks (DMINs), rather than by enlargement of the diffusion species caused by molecular aggregation. Constructed through the synergistic interactions of various polar molecules in crude oil, DMINs dynamically bind polar molecules, trap polarizable molecules, and spatially hinder the free motion of non-polar molecules. Overall, this reduces the mobility of all molecular species, as illustrated by the decreased diffusion coefficients. This study demonstrates that DOSY is a powerful NMR method to investigate molecular motion abilities also in complex mixtures. In addition, the insights in the influence of the interaction matrix on the molecular mobility also help to understand the contribution of "structural viscosity" to the viscosity of heavy oil.

Spatial-Spectral Constrained Adaptive Graph for Hyperspectral Image Clustering.

Hyperspectral image (HSI) clustering is a challenging task, whose purpose is to assign each pixel to a corresponding cluster. The high-dimensionality and noise corruption are two main problems that limit the performance of HSI clustering. To address those problems, this paper proposes a projected clustering with a spatial-spectral constrained adaptive graph (PCSSCAG) method for HSI clustering. PCSSCAG first constructs an adaptive adjacency graph to capture the accurate local geometric structure of HSI data adaptively. Then, a spatial-spectral constraint is employed to simultaneously explore the spatial and spectral information for reducing the negative influence on graph construction caused by noise. Finally, projection learning is integrated into the spatial-spectral constrained adaptive graph construction for reducing the redundancy and alleviating the computational cost. In addition, an alternating iteration algorithm is designed to solve the proposed model, and its computational complexity is theoretically analyzed. Experiments on two different scales of HSI datasets are conducted to evaluate the performance of PCSSCAG. The associated experimental results demonstrate the superiority of the proposed method for HSI clustering.

Correction: Towards high-throughput microfluidic Raman-activated cell sorting.

Correction for 'Towards high-throughput microfluidic Raman-activated cell sorting' by Qiang Zhang et al., Analyst, 2015, DOI: 10.1039/c5an01074h.

Towards high-throughput microfluidic Raman-activated cell sorting.

Raman-activated cell sorting (RACS) is a promising single-cell analysis technology that is able to identify and isolate individual cells of targeted type, state or environment from an isogenic population or complex consortium of cells, in a label-free and non-invasive manner. However, compared with those widely used yet labeling-required or staining-dependent cell sorting technologies such as FACS and MACS, the weak Raman signal greatly limits the further development of the existing RACS systems to achieve higher throughput. Strategies that can tackle this bottleneck include, first, improvement of Raman-acquisition efficiency and quality based on advanced Raman spectrometers and enhanced Raman techniques; second, development of novel microfluidic devices for cell sorting followed by integration into a complete RACS system. Exploiting these strategies, prototypes for a new generation of RACS have been demonstrated, such as flow-based OT-RACS, DEP-RACS, and SERS/CARS flow cytometry. Such high-throughput microfluidic RACS can provide biologists with a powerful single-cell analysis tool to explore the scientific questions or applications that have been beyond the reach of FACS and MACS.

An exceptional kinetic quantum sieving separation effect of hydrogen isotopes on commercially available carbon molecular sieves.

The quantum sieving effect of H2/D2 at 77 K on commercially available carbon molecular sieves (1.5GN-H and 3KT-172) was studied. An exceptional reverse kinetic quantum effect is observed on 1.5GN-H where D2 diffuses much faster than H2 with a ratio of up to 5.83 at low pressure, and the difference is still very evident even as the pressure increases up to 1 bar. D2 also diffuses faster than H2 on 3KT-172 with a ratio of up to 1.86. However, the reverse kinetic sieving disappears in a polymer-based carbon (PC). The present kinetic quantum sieving effect of H2 and D2 at 77 K on 1.5GN-H is the highest to date.

What makes the unsupervised monocular depth estimation (UMDE) model training better.

Current computer vision tasks based on deep learning require a huge amount of data with annotations for model training or testing, especially in some dense estimation tasks, such as optical flow segmentation and depth estimation. In practice, manual labeling for dense estimation tasks is very difficult or even impossible, and the scenes of the dataset are often restricted to a small range, which dramatically limits the development of the community. To overcome this deficiency, we propose a synthetic dataset generation method to obtain the expandable dataset without burdensome manual workforce. By this method, we construct a dataset called MineNavi containing video footages from first-perspective-view of the aircraft matched with accurate ground truth for depth estimation in aircraft navigation application. We also provide quantitative experiments to prove that pre-training via our MineNavi dataset can improve the performance of depth estimation model and speed up the convergence of the model on real scene data. Since the synthetic dataset has a similar effect to the real-world dataset in the training process of deep model, we finally conduct the experiments on MineNavi with unsupervised monocular depth estimation (UMDE) deep learning models to demonstrate the impact of various factors in our dataset such as lighting conditions and motion mode, aiming to explore what makes this kind of models training better.

Self-assembled functional organic monolayers on oxide-free copper.

The preparation and characterization of self-assembled monolayers on copper with n-alkyl and functional thiols was investigated. Well-ordered monolayers were obtained, while the copper remained oxide-free. Direct attachment of N-succinimidyl mercaptoundecanoate (NHS-MUA) onto the copper surface allowed for the successful attachment of biomolecules, such as ß-d-glucosamine, the tripeptide glutathione, and biotin. Notably, the copper surfaces remained oxide-free even after two reaction steps. All monolayers were characterized by static water contact angle measurements, X-ray photoelectron spectroscopy, and infrared reflection absorption spectroscopy. In addition, the biotinylated copper surfaces were employed in the immobilization of biomolecules such as streptavidin.

RLStereo: Real-Time Stereo Matching Based on Reinforcement Learning.

Many state-of-the-art stereo matching algorithms based on deep learning have been proposed in recent years, which usually construct a cost volume and adopt cost filtering by a series of 3D convolutions. In essence, the possibility of all the disparities is exhaustively represented in the cost volume, and the estimated disparity holds the maximal possibility. The cost filtering could learn contextual information and reduce mismatches in ill-posed regions. However, this kind of methods has two main disadvantages: 1) cost filtering is very time-consuming, and it is thus difficult to simultaneously satisfy the requirements for both speed and accuracy; 2) thickness of the cost volume determines the disparity range which can be estimated, and the pre-defined disparity range may not meet the demand of practical application. This paper proposes a novel real-time stereo matching method called RLStereo, which is based on reinforcement learning and abandons the cost volume or the routine of exhaustive search. The trained RLStereo makes only a few actions iteratively to search the value of the disparity for each pair of stereo images. Experimental results show the effectiveness of the proposed method, which achieves comparable performances to state-of-the-art algorithms with real-time speed on the public large-scale testset, i.e., Scene Flow.

Room-temperature intermediate layer bonding for microfluidic devices.

In this work a novel room-temperature bonding technique based on chemically activated Fluorinated Ethylene Propylene (FEP) sheet as an intermediate between chemically activated substrates is presented. Surfaces of silicon and glass substrates are chemically modified with APTES bearing amine terminal groups, while FEP sheet surfaces are treated to form carboxyl groups and subsequently activated by means of EDC-NHS chemistry. The activation procedures of silicon, glass and FEP sheet are characterized by contact angle measurements and XPS. Robust bonds are created at room-temperature by simply pressing two amine-terminated substrates together with activated FEP sheet in between. Average tensile strengths of 5.9 MPa and 5.2 MPa are achieved for silicon-silicon and glass-glass bonds, respectively, and the average fluidic pressure that can be operated is 10.2 bar. Moreover, it is demonstrated that FEP-bonded microfluidic chips can handle mild organic solvents at elevated pressures without leakage problems. This versatile room-temperature intermediate layer bonding technique has a high potential for bonding, packaging, and assembly of various (bio-) chemical microfluidic systems and MEMS devices.

Kinetics of atomic force microscope-based scanned probe oxidation on an octadecylated silicon(111) surface.

Atomic force microscope (AFM)-based scanned probe oxidation (SPO) nanolithography has been carried out on an octadecyl-terminated Si(111) surface to create dot-array patterns under ambient conditions in contact mode. The kinetics investigations indicate that this SPO process involves three stages. Within the steadily growing stage, the height of oxide dots increases logarithmically with pulse duration and linearly with pulse voltage. The lateral size of oxide dots tends to vary in a similar way. Our experiments show that a direct-log kinetic model is more applicable than a power-of-time law model for the SPO process on an alkylated silicon in demonstrating the dependence of oxide thickness on voltage exposure time within a relatively wide range. In contrast with the SPO on the octodecysilated SiO2/silicon surface, this process can be realized by a lower voltage with a shorter exposure time, which will be of great benefit to the fabrication of integrated nanometer-sized electronic devices on silicon-based substrates. This study demonstrates that the alkylated silicon is a new promising substrate material for silicon-based nanolithography.

L1-Norm Low-Rank Matrix Decomposition by Neural Networks and Mollifiers.

The L1-norm cost function of the low-rank approximation of the matrix with missing entries is not smooth, and also cannot be transformed into a standard linear or quadratic programming problem, and thus, the optimization of this cost function is still not well solved. To tackle this problem, first, a mollifier is used to smooth the cost function. High closeness of the smoothed function to the original one can be obtained by tuning the parameters contained in the mollifier. Next, a recurrent neural network is proposed to optimize the mollified function, which will converge to a local minimum. In addition, to boost the speed of the system, the mollifying process is implemented by a filtering procedure. The influence of two mollifier parameters is theoretically analyzed and experimentally confirmed, showing that one of the parameters is critical to computational efficiency and accuracy, while the other not. A large number of experiments on synthetic data show that the proposed method is competitive to the state-of-the-art methods. In particular, the experiments on large matrices and a real application in the structure from motion indicate that the memory requirement of the proposed algorithm is mild, making it suitable for real applications that often involve large-scale matrix decomposition.

Capillary ion chromatography-mass spectrometry for simultaneous determination of glucosylglycerol and sucrose in intracellular extracts of cyanobacteria.

A capillary ion chromatography-mass spectrometry (MS) method was proposed to determine glucosylglycerol (GG), sucrose, and five other carbohydrates. MS conditions and make-up flow parameters were optimized. This method is accurate and sensitive for simultaneous analysis of carbohydrates, with mean correlation coefficients of determination greater than 0.99, relative standard deviation of 0.91-2.81% for eight replicates, and average spiked recoveries of 97.3-104.9%. Limits of detection of sodium adduct were obtained with MS detection in selected ion mode for GG (0.006mg/L), sucrose (0.02mg/L), and other carbohydrates (0.03mg/L). This method was successfully applied to determine GG and sucrose in intracellular extracts of salt-stressed cyanobacteria.

On-demand control of microfluidic flow via capillary-tuned solenoid microvalve suction.

A simple, low-cost and on-demand microfluidic flow controlling platform was developed based on a unique capillary-tuned solenoid microvalve suction effect without any outer pressure source. The suction effect was innovatively employed as a stable and controllable driving force for the manipulation of the microfluidic system by connecting a piece of capillary between the microvalve and the microfluidic chip, which caused significant hydrodynamic resistance differences among the solenoid valve ports and changed the flowing mode inside the valve. The volume of sucked liquid could be controlled from microliters even down to picoliters either by decreasing the valve energized duration (from a maximum energized duration to the valve response time of 20 ms) or by increasing the inserted capillary length (i.e., its hydrodynamic resistance). Several important microfluidic unit operations such as cell/droplet sorting and on-demand size-controllable droplet generation have been demonstrated on the developed platform and both simulations and experiments confirmed that this platform has good controllability and stability.

Quantitative analysis of fatty-acid-based biofuels produced by wild-type and genetically engineered cyanobacteria by gas chromatography-mass spectrometry.

Simple and rapid quantitative determination of fatty-acid-based biofuels is greatly important for the study of genetic engineering progress for biofuels production by microalgae. Ideal biofuels produced from biological systems should be chemically similar to petroleum, like fatty-acid-based molecules including free fatty acids, fatty acid methyl esters, fatty acid ethyl esters, fatty alcohols and fatty alkanes. This study founded a gas chromatography-mass spectrometry (GC-MS) method for simultaneous quantification of seven free fatty acids, nine fatty acid methyl esters, five fatty acid ethyl esters, five fatty alcohols and three fatty alkanes produced by wild-type Synechocystis PCC 6803 and its genetically engineered strain. Data obtained from GC-MS analyses were quantified using internal standard peak area comparisons. The linearity, limit of detection (LOD) and precision (RSD) of the method were evaluated. The results demonstrated that fatty-acid-based biofuels can be directly determined by GC-MS without derivation. Therefore, rapid and reliable quantitative analysis of fatty-acid-based biofuels produced by wild-type and genetically engineered cyanobacteria can be achieved using the GC-MS method founded in this work.

Light-enhanced microcontact printing of 1-alkynes onto hydrogen-terminated silicon.

A method for the direct patterning of 1-alkynes onto hydrogen-terminated silicon is presented. It combines microcontact printing with illumination through the stamp, and results in the formation of an alkenyl monolayer. The formation of heterogeneous monolayers is demonstrated by subsequent backfilling.

Local probe oxidation of self-assembled monolayers on hydrogen-terminated silicon.

Local probe oxidation experiments by conductive AFM have been performed on a hexadecyl monolayer and a N-hydroxysuccinimide (NHS)-ester-functionalized undecyl (NHS-UA) monolayer assembled on hydrogen-terminated (i.e., unoxidized) silicon. The oxidation conditions for the mild oxidation of the top terminal groups of monolayers and the deep oxidation of the underlying silicon into silicon oxide were investigated. The results show that the bias threshold for the AFM tip-induced oxidation of the top groups of monolayers on oxide-free silicon can be reduced by 2 V for the methyl-terminated hexadecyl monolayer and even by 3.5 V for the active NHS-ester-terminated undecyl monolayer, in comparison to a methyl-terminated octadecyl trichlorosilane (OTS) monolayer on oxidized silicon. Upon such local mild oxidation, the active NHS ester group of the NHS-UA monolayer is selectively cleaved off to generate carboxyl-containing monolayer nanopatterns, opening further possibilities for subsequent patterned multifunctionalization.

Photochemical covalent attachment of alkene-derived monolayers onto hydroxyl-terminated silica.

The functionalization of optically transparent substrates is of importance, for example, in the field of biosensing. In this article, a new method for modification of silica surfaces is presented that is based on a photochemical reaction of terminal alkenes with the surface. This yields highly hydrophobic surfaces, which are thermally stable up to at least 400 degrees C. The formed monolayer provides chemical passivation of the underlying surface, according to studies showing successful blocking of platinum atomic layer deposition (ALD). The reaction is photochemically initiated, requiring light with a wavelength below 275 nm. X-ray photoelectron spectroscopy and infrared spectroscopy studies show that the alkenes initially bind to the surface hydroxyl groups in Markovnikov fashion. At prolonged reaction times (>5 h), however, oligomerization occurs, resulting in layer growth normal to the surface. The photochemical nature of the reaction enables the use of photolithography as a tool to constructively pattern silica surfaces. Atomic force microscopy shows that the features of the photomask are well transferred. The newly developed method can complement existing patterning methods on silica that are based on soft lithography.

One-step photochemical attachment of NHS-terminated monolayers onto silicon surfaces and subsequent functionalization.

N-Hydroxysuccinimide (NHS)-ester-terminated monolayers were covalently attached in one step onto silicon using visible light. This mild photochemical attachment, starting from omega-NHS-functionalized 1-alkenes, yields a clean and flat monolayer-modified silicon surface and allows a mild and rapid functionalization of the surface by substitution of the NHS-ester moieties with amines at room temperature. Using a combination of analytical techniques (infrared reflection absorption spectroscopy (IRRAS), extensive X-ray photoelectron spectroscopy (XPS) in combination with density functional theory calculations of the XPS chemical shifts of the carbon atoms, atomic force microscopy (AFM), and static contact angle measurements), it was shown that the NHS-ester groups were attached fully intact onto the surface. The surface reactivity of the NHS-ester moieties toward amines was qualitatively and quantitatively evaluated via the reaction with para-trifluoromethyl benzylamine and biotin hydrazide.