RailFOD23: A dataset for foreign object detection on railroad transmission lines.

Artificial intelligence models play a crucial role in monitoring and maintaining railroad infrastructure by analyzing image data of foreign objects on power transmission lines. However, the availability of publicly accessible datasets for railroad foreign objects is limited, and the rarity of anomalies in railroad image data, combined with restricted data sharing, poses challenges for training effective foreign object detection models. In this paper, the aim is to present a new dataset of foreign objects on railroad transmission lines, and evaluating the overall performance of mainstream detection models in this context. Taking a unique approach and leveraging large-scale models such as ChatGPT (Chat Generative Pre-trained Transformer) and text-to-image generation models, we synthesize a series of foreign object data. The dataset includes 14,615 images with 40,541 annotated objects, covering four common foreign objects on railroad power transmission lines. Through empirical research on this dataset, we validate the performance of various baseline models in foreign object detection, providing valuable insights for the monitoring and maintenance of railroad facilities.

Anisotropic Interfacial Force Field for Interfaces of Water with Hexagonal Boron Nitride.

This study introduces an anisotropic interfacial potential that provides an accurate description of the van der Waals (vdW) interactions between water and hexagonal boron nitride (h-BN) at their interface. Benchmarked against the strongly constrained and appropriately normed functional, the developed force field demonstrates remarkable consistency with reference data sets, including binding energy curves and sliding potential energy surfaces for various configurations involving a water molecule adsorbed atop the h-BN surface. These findings highlight the significant improvement achieved by the developed force field in empirically describing the anisotropic vdW interactions of the water/h-BN heterointerfaces. Utilizing this anisotropic force field, molecular dynamics simulations demonstrate that atomically flat, pristine h-BN exhibits inherent hydrophobicity. However, when atomic-step surface roughness is introduced, the wettability of h-BN undergoes a significant change, leading to a hydrophilic nature. The calculated water contact angle (WCA) for the roughened h-BN surface is approximately 64°, which closely aligns with experimental WCA values ranging from 52° to 67°. These findings indicate the high probability of the presence of atomic steps on the surfaces of the experimental h-BN samples, emphasizing the need for further experimental verification. The development of the anisotropic interfacial force field for accurately describing interactions at the water/h-BN heterointerfaces is a significant advancement in accurately simulating the wettability of two-dimensional (2D) materials, offering a reliable tool for studying the dynamic and transport properties of water at these interfaces, with implications for materials science and nanotechnology.

Federated Transfer Learning for Low-dose PET Denoising: A Pilot Study with Simulated Heterogeneous Data.

Positron emission tomography (PET) with a reduced injection dose, i.e., low-dose PET, is an efficient way to reduce radiation dose. However, low-dose PET reconstruction suffers from a low signal-to-noise ratio (SNR), affecting diagnosis and other PET-related applications. Recently, deep learning-based PET denoising methods have demonstrated superior performance in generating high-quality reconstruction. However, these methods require a large amount of representative data for training, which can be difficult to collect and share due to medical data privacy regulations. Moreover, low-dose PET data at different institutions may use different low-dose protocols, leading to non-identical data distribution. While previous federated learning (FL) algorithms enable multi-institution collaborative training without the need of aggregating local data, it is challenging for previous methods to address the large domain shift caused by different low-dose PET settings, and the application of FL to PET is still under-explored. In this work, we propose a federated transfer learning (FTL) framework for low-dose PET denoising using heterogeneous low-dose data. Our experimental results on simulated multi-institutional data demonstrate that our method can efficiently utilize heterogeneous low-dose data without compromising data privacy for achieving superior low-dose PET denoising performance for different institutions with different low-dose settings, as compared to previous FL methods.

FedFTN: Personalized federated learning with deep feature transformation network for multi-institutional low-count PET denoising.

Low-count PET is an efficient way to reduce radiation exposure and acquisition time, but the reconstructed images often suffer from low signal-to-noise ratio (SNR), thus affecting diagnosis and other downstream tasks. Recent advances in deep learning have shown great potential in improving low-count PET image quality, but acquiring a large, centralized, and diverse dataset from multiple institutions for training a robust model is difficult due to privacy and security concerns of patient data. Moreover, low-count PET data at different institutions may have different data distribution, thus requiring personalized models. While previous federated learning (FL) algorithms enable multi-institution collaborative training without the need of aggregating local data, addressing the large domain shift in the application of multi-institutional low-count PET denoising remains a challenge and is still highly under-explored. In this work, we propose FedFTN, a personalized federated learning strategy that addresses these challenges. FedFTN uses a local deep feature transformation network (FTN) to modulate the feature outputs of a globally shared denoising network, enabling personalized low-count PET denoising for each institution. During the federated learning process, only the denoising network's weights are communicated and aggregated, while the FTN remains at the local institutions for feature transformation. We evaluated our method using a large-scale dataset of multi-institutional low-count PET imaging data from three medical centers located across three continents, and showed that FedFTN provides high-quality low-count PET images, outperforming previous baseline FL reconstruction methods across all low-count levels at all three institutions.

An Intelligent Waste-Sorting and Recycling Device Based on Improved EfficientNet.

The main source of urban waste is the daily life activities of residents, and the waste sorting of residents' waste is important for promoting economic recycling, reducing labor costs, and protecting the environment. However, most residents are unable to make accurate judgments about the categories of household waste, which severely limits the efficiency of waste sorting. We have designed an intelligent waste bin that enables automatic waste sorting and recycling, avoiding the extensive knowledge required for waste sorting. To ensure that the waste-classification model is high accuracy and works in real time, GECM-EfficientNet is proposed based on EfficientNet by streamlining the mobile inverted bottleneck convolution (MBConv) module, introducing the efficient channel attention (ECA) module and coordinate attention (CA) module, and transfer learning. The accuracy of GECM-EfficientNet reaches 94.54% and 94.23% on the self-built household waste dataset and TrashNet dataset, with parameters of only 1.23 M. The time of one recognition on the intelligent waste bin is only 146 ms, which satisfies the real-time classification requirement. Our method improves the computational efficiency of the waste-classification model and simplifies the hardware requirements, which contributes to the residents' waste classification based on intelligent devices.

Spontaneous Movement of a Droplet on a Conical Substrate: Theoretical Analysis of the Driving Force.

Experiments and simulations have shown that a droplet can move spontaneously and directionally on a conical substrate. The driving force originating from the gradient of curvatures is revealed as the self-propulsion mechanism. Theoretical analysis of the driving force is highly desirable; currently, most of them are based on a perturbative theory with assuming a weakly curved substrate. However, this assumption is valid only when the size of the droplet is far smaller than the curvature radius of the substrate. In this paper, we derive a more accurate analytical model for describing the driving force by exploring the geometric characteristics of a spherical droplet on a cylindrical substrate. In contrast to the perturbative solution, our model is valid under a much weaker condition, i.e., the contact region between the droplet and the substrate is small compared with the curvature radius of the substrate. Therefore, we show that for superhydrophobic surfaces, the derived analytical model is applicable even if the droplet is very close to the apex of a conical substrate. Our approach opens an avenue for studying the behavior of droplets on the tip of the conical substrate theoretically and could also provide guidance for the experimental design of curved surfaces to control the directional motion of droplets.

Network Activity Due to Topographic Organization of Schaffer Collaterals in a Large-Scale Model of Rat CA1.

Connectivity between neural regions, particularly in the hippocampus, is seldom all-to-all or random, yet it is the predominant method by which connectivity is implemented in most models of neuronal networks. We have been developing a computational platform for simulating the trisynaptic circuit of rat hippocampus with which we have constructed a large-scale, biologically-realistic, spiking neuronal network model of the entorhinal-dentate-CA3 system. Using the model, we had demonstrated a non-trivial effect of topographic connectivity on network dynamics and function. In this work, we detail the introduction of the CA1 subregion to the large-scale model. Using anatomical data, we constrained the distribution of axon collaterals, i.e., Schaffer collaterals, projected from CA3 to CA1 and preserved the topographic organization of the projections. Using a simplified multi-compartmental model of CA1 pyramidal cells and a single compartment model of CA1 parvalbumin basket cells, that were connected with disynaptic feedforward inhibition and feedback inhibition, we demonstrate the network activity of the CA1 network given a topographic organization of Schaffer collaterals. From this introduction of CA1 to the large-scale model, we can then observe the successive transformation of spatio-temporal, spiking neural activity as it propagates through the trisynaptic circuit.

Chitosan-based membrane chromatography for protein adsorption and separation.

A chitosan-based membrane chromatography was set up by using natural chitosan/carboxymethylchitosan (CS/CMCS) blend membrane as the matrix. The dynamic adsorption property for protein (lysozyme as model protein) was detailed discussed with the change in pore size of the membrane, the flow rate and the initial concentration of the feed solution, and the layer of membrane in membrane stack. The best dynamic adsorption capacity of lysozyme on the CS/CMCS membrane chromatography was found to be 15.3mg/mL under the optimal flow conditions. Moreover, the CS/CMCS membrane chromatography exhibited good repeatability and reusability with the desorption efficiency of ~90%. As an application, lysozyme and ovalbumin were successfully separated from their binary mixture through the CS/CMCS membrane chromatography. This implies that such a natural chitosan-based membrane chromatography may have great potential on the bioseparation field in the future.

[Research on source profile of aerosol organic compounds in leather plant].

Through investigating current air pollution condition for PM10 in every factories of different style leather plants in Pearl River Delta, characteristic profile of semi-volatile organic compounds in PM10 emitted from leather factories and their contents were researched by using ultrasonic and gas chromatography and mass spectrum technology. The 6 types of organic compounds containing 46 species in total were found in the collected samples, including phenyl compounds, alcohols, PAHs, acids, esters and amides. The concentrations of PM10 in leather tanning plant, leather dying plant and man-made leather plant were 678.5, 454.5, 498.6 microgm x m(-3) respectively, and concentration of organic compounds in PM10 were 10.04, 6.89, 14.21 microg x m(-3) in sequence. The more important type of pollutants in each leather plants had higher contribution to total organic mass as follows, esters and amides in tanning plants profile account for 43.47% and 36.51% respectively; esters and alcohols in dying plants profiles account for 52.52% and 16.16% respectively; esters and amide in man-made leather plant have the highest content and account for 57.07% and 24.17% respectively. In the aerosol organic source profiles of tested leather plants, 9-octadecenamide was the abundant important species with the weight of 26.15% in tanning plant, and Bis(2-ethylhexyl) phthalate was up to 44.19% in the dying plant, and Bis(2-ethylhexyl) maleate and 1-hydroxy-piperidine had obviously higher weight in man-made plant than the other two plants.

Protein adsorption and separation on amphoteric chitosan/carboxymethylcellulose membranes.

This article reported the preparation of an amphoteric natural polymeric membrane-macroporous chitosan (CS)/carboxymethylcellulose (CMC) blend membrane and the utilization of such a membrane on the membrane chromatography for bioseparation. The membranes were prepared by solution blending of CS and CMC solution, and using silica particles as porogen. Both glutaraldehyde and epichlorohydrin were used as crosslinking agent to increase its chemical stability in aqueous solution. Such a natural polymeric membrane can be served as an amphoteric membrane because of the amino group on CS and the carboxymethyl group on CMC, in which the surface charge can be changed with the environmental pH. Ovalbumin (pI = 4.6) and lysozyme (pI = 11) were selected as model proteins. These two proteins adsorption on different CS/CMC blend membranes with different initial protein concentrations at different pH values were investigated in batch systems. The results indicated that the maximum adsorption for lysozyme and ovalbumin was at pH 9.2 and 4.8 respectively, and the adsorption capacity on the membrane both increased with the increase of initial protein concentration. Though the adsorption mechanism of lysozyme and ovalbumin was found not the same, the maximum adsorption capacity of two proteins on the membranes was quite similar (about 250 mg/g). Moreover, the desorption ratio of both proteins was found to be more than 90% that implied CS/CMC blend membrane could separate proteins by adsorption-desorption process. Finally, both lysozyme and ovalbumin were successfully separated from their binary mixture only by adjusting the pH of the feed and the desorption solution.