Regime shifts and their underlying mechanisms in ecosystems are a critical issue in ecological research， with profound implications for predicting ecological risks under global change. This review systematically synthesizes the theoretical advances in alternative stable state （ASS） of ecosystem multi-stability， focusing on three key dimensions： micro-macro process coupling， mechanisms of threshold response， and the role of regulatory nodes in ecosystem resilience. By integrating methodologies such as ASS theory， potential landscape modeling， and bifurcation analysis， we highlight how climate change and anthropogenic activities are driving critical ecosystems （e.g.， coral reefs， Amazon rainforest， Arctic permafrost） toward tipping points， while hysteresis effects and irreversible potentials exacerbate recovery challenges. Emerging approaches combining network theory and energy （carbon） flux analysis offer novel insights for cross-scale early warning， yet bridging micro-scale mechanisms with macro-scale patterns remains a critical challenge. This review provides a theoretical framework for ecological threshold management and underscores the urgent need for interdisciplinary approaches to address planetary-scale regime shift risks.

Semi-supervised learning is a key research problem in the field of pattern recognition and machine learning， and has been widely used in various fields in recent years. In practical problems， labeled samples are costly to obtain， while unlabeled samples are easier to obtain despite the lack of labeling information. Semi-supervised learning uses a large amount of unlabeled data and a small amount of labeled data at the same time to perform pattern recognition work. In this paper， we propose a robust semi-supervised approach based on model averaging and \gamma-divergence： on the one hand， the problem of low quality of unlabeled data is addressed by introducing model averaging method； on the other hand， the problem of mislabeling of labeled data is addressed by introducing logistic regression based on \gamma-divergence. One of the advantages of the proposed model is that we can process the data by exploiting the predictive differences among the different models to effectively utilize the information of the unlabeled data while minimizing the harmful information in it. By introducing \gamma-divergence to reduce the effect of mislabeled data in labeled data on the fitting effect， we ultimately obtain a model that is robust for both unlabeled and labeled data. Simulation studies and applications of Breast Cancer Wisconsin Dataset show that compared with existing semi-supervised learning methods， the new method proposed in this paper has a significant improvement in prediction performance when the data quality is low.

The cylindrical structure finds extensive applications in ocean engineering， where the shedding of wake vortices can generate fluctuating drag that may induce vortex-induced vibrations， which can lead to equipment instability. Reducing the flow drag around the cylinder can help suppress vortex-induced vibrations. In recent years， the use of cactus-inspired riblets and superhydrophobic coatings has emerged as two efficient passive control strategies for drag reduction. This study combines these two methods to conduct experiments on drag reduction in underwater cylindrical flow. The experimental results demonstrate that surface riblets can decrease the drag on the cylinder， and the addition of superhydrophobic surfaces can further enhance drag reduction. Surface riblets extend the recirculation region while superhydrophobic surfaces shorten it. By utilizing Proper Orthogonal Decomposition to compute the phase-averaged flow field， it is found that， compared to hydrophilic cylinders， superhydrophobic surfaces lead to a reduction in the cylinder’s wake shear layer and cause the vortex to form closer to the cylinder， thereby shortening the recirculation region. Overall， the combination of surface riblets and superhydrophobic coatings effectively reduces the flow resistance around the cylinder， offering potential applications in flow control and drag reduction technologies.

This study investigates the heat transfer characteristics of the Toco Toucan’s beak， which is known for its unique structural features and strong heat exchange capabilities， using natural convection numerical simulations in environments of 30 ℃ and 15 ℃， respectively. Temperature contours at different positions along the length of the beak were extracted. It was observed that the heat transfer efficiency of the beak is higher in high-temperature environments， whereas in low-temperature environments， only a distinct temperature boundary layer near the skull is evident. Analysis revealed significant variations in the local Rayleigh number （Ra_x ） near the skull in low-temperature environments， while Ra_x in the anterior midsection of the beak remains relatively small， resulting in less pronounced convective heat exchange in this region. Streamline diagrams illustrate that in high-temperature environments， the entrainment effect at the tip of the beak alleviates the heat exchange deficiency caused by the small surface area， effectively utilizing every part of the beak’s dissipating surface. However， in low-temperature environments， the entrainment effect of the beak is concentrated near the skull， resulting in inevitable heat loss. By analyzing three dimensionless numbers， C_p， C_f， and Nu， it was found that C_p values in the Maxilla are negative in both environments， promoting the influx of cold air into the boundary layer and improving heat exchange efficiency by reducing temperature differentials caused by preheating effects. Particularly in low-temperature environments， C_p and C_{fvalues in the anterior midsection of the beak are almost zero， while Nu stabilizes at a relatively small value， minimizing heat loss from the beak’s surface. The above research results quantitatively elucidated the heat exchange characteristics of bird beaks. Through further studies， it is hoped to provide reference for exploring the geographical distribution of toucans.}

Liquid metal， as a highly efficient heat transport medium， is very important in the design of magnetic confinement fusion devices. Predicting the solidification law of liquid metal under strong magnetic fields is an important issue. This article uses a high-speed camera to capture the phenomenon of gallium indium tin alloy impacting metal solid walls and solidifying under the action of a vertical magnetic field. Using image processing technology， it summarizes the characteristics of metal droplet impact spreading， rebound， and solidification under different undercooling temperatures， impact velocities， and magnetic induction intensity ranges of the bottom plate. The bottom plate temperature is 20， -30， -40， and -50 ℃， the impact velocity is 0.45-1.71 m/s， and the magnetic field strength range is 0-1.2 T. Experimental phenomena show that when droplets impact isothermal walls and supercooled walls， the dimensionless scaling rate of the maximum spreading factor follows the classical theoretical prediction relationship. The magnetic field initially promotes and then suppresses the rebound height of droplets. The empirical relationship between the maximum spreading factor and N under the action of magnetic field is derived. The magnetic field inhibits the separation of droplets when they hit the supercooled wall， and weakens the oscillation in the height direction to promote solidification.

The use of excavated soil from highway construction sites to produce foamed polymeric soil， employed as a thermal insulation layer for seasonally frozen subgrades， not only achieves comprehensive resource utilization but also mitigates frost damage to the subgrades. Based on the excavated soil along the Urumqi Ring Expressway， combined with cement and foam， the foamed polymeric soil with porous insulation properties was developed. The influences of wet density， soil admixture， and water-cement ratio on the thermal conductivity， unconfined compressive strength， and stiffness of foamed polymeric soil were systematically investigated， establishing an intrinsic connection between pore structure and macroscopic performance. The research findings indicate that when the density of foamed polymeric soil increases from 600 kg/m³ to 1 200 kg/m³， its thermal conductivity approximately doubles， strength increases by about 3.95 times， and modulus increases by approximately 10.5 times. Compared to traditional subgrade soil， the thermal conductivity of foamed polymeric soil is significantly reduced by 52%-96.4%. Further analysis of the pore structure reveals that as the pore size of foamed polymeric soil within a unit volume decreases， the proportion of the skeleton increases correspondingly， while the volume of air pores decreases. This microstructural change manifests as improved thermal insulation performance （i.e.， reduced thermal conductivity） and significant enhancement in mechanical properties （including strength and stiffness） at the macroscopic level. Foamed polymeric soil not only demonstrates better thermal insulation performance but also exhibits strong mechanical characteristics， providing a suitable solution for the thermal insulation layer of seasonally frozen subgrades.

The medium-earth-orbit synthetic aperture radar （MEO SAR） possesses the capability of short revisit period， high spatial resolution， and wide coverage， which has promising applications in both military and civilian fields. The MEO SAR can improve azimuth resolution by operating in sliding spotlight mode， and can enhance system flexibility through squint imaging mode. However， the increase in orbit height invalidates the assumption of azimuth invariance and the squint mode causes spectrum distortion. In this paper， first， we derive the calculation process of imaging parameters based on the polynomial signal model. Then， by introducing a new model fitting method and a deramp function， we extend the azimuth time-domain resampling method from the stripmap mode to the sliding spotlight mode， correcting the azimuth space-variant characteristics. We further propose a processing flow by combining the two-step method and the non-linear range walk correction to solve the spectrum aliasing in the squint sliding spotlight mode. Finally， the proposed method is verified through simulation experiments.

As the resolution of airborne synthetic aperture radar continues to improve， motion compensation has become a core link to ensure high-quality imaging. This paper conducts an in-depth analysis of the motion error of chirped pulse SAR， and proposes a Chirplet transform motion compensation algorithm based on original echo data to address the problem of non-space-variant phase error compensation of SAR. This algorithm uses Chirplet transform for time-frequency analysis to accurately characterize the phase error in the echo data， uses maximum likelihood estimation to extract the modulation frequency of the phase error， and solves the phase error for motion compensation. Compared with the PGA algorithm， this algorithm does not need to rely on strong scattering points and can achieve better image focusing performance in the presence of large phase errors. Compared with the MD subaperture algorithm， this algorithm performs frequency modulation rate estimation for each point， resulting in a more precise estimation. Finally， simulation experiments and quantitative analysis verified the effectiveness of the algorithm.

Super-resolution technology has become an important tool for reconstructing high-resolution datasets and supplementing the shortage of high-resolution images with its characteristics of flexibility and low cost. Compared with natural images， remote sensing images of real scenes are complex and specific， which make super-resolution tasks more difficult. Meanwhile， for remote sensing images， traditional deep learning models can improve the resolution， but there is still a great deficiency of improvement for the details and textures of the ground objects. Therefore， based on the generative adversarial network model， this paper fuses channel-space attention to enhance the feature learning capability of the network and use an artifact suppression strategy to distinguish smooth regions from detail-rich regions， so that the network can focus more on detail-rich regions and suppress the generation of artifacts. Extensive experiments on GaoFen satellite data show that the quantitative metrics and visual quality of the method proposed in this paper are better than those of the current mainstream methods.

Single image super-resolution （SISR） can improve the resolution of remote sensing images （RSIs）， thereby improving the application value of data. At present， the number of pixels of RSIs generally reaches hundreds of millions， and it is usually necessary to divide the image into patches when performing SISR. However， there is a lack of relevant research on how to effectively determine the patch size and whether different sizes affect the results. In this paper， taking a large-scale high-resolution RSIs as the experiment data， 3 typical SISR models are selected， 9 groups of SR experiments under different patch sizes are carried out， and the super-resolution （SR） results for the whole of the large-scaled RSI are analyzed comprehensively both qualitatively and quantitatively. The results show that： 1） Cutting of the patches results in stitching seams at the stitching place. In particular， when the patch size is small， a large number of stitching seams show a block effect and the inconsistency is more obvious. 2） With the increase of the patch size， the SR accuracy of the three models is improved， and the overall computational efficiency is also improved. When the test patches are larger than the training patches， the elapsed time and accuracy stabilize. 3） The feasibility and accuracy of the whole RSI input are closely related to the model. The ESPCN model has the best accuracy when inputting the whole RSI， the RDBPN model may cause the accuracy to decrease due to the non-square matrix of the RSI， and the HSENET model has high requirements for computing power and cannot calculate the whole RSI. In conclusion， this paper provides an experimental basis for the selection of patch size for RSI SR engineering applications.

Huanjing-1A （HJ-1A） CCD1 has four reflected solar bands in visible and near-infrared bands. The on-orbit absolute radiometric calibration frequency of HJ-1A/CCD1 is limited to the annual site calibration. Cross calibration， as a supplement means of site calibration， can achieve high-frequency on-orbit absolute radiometric calibration. In this paper， we select Terra/MODIS as the reference sensor to perform long time series cross calibration on HJ-1A/CCD1 based on Dunhuang calibration site. By limiting conditions such as the imaging angle difference （less than 20°）， the imaging time difference （less than 2 hours）， cloud cover， and imaging quality between HJ-1A/CCD1 and Terra/MODIS， a total of 147 effective cross calibration image pairs were screened out from September 2008 to December 2021. Then the 6SV v2.1 radiative transfer model was used to calculate the spectral band adjustment factor， and ultimately achieving long time series cross calibration of HJ-1A/CCD1. The results show that： 1） The cross-calibration coefficients calculated in this article are highly consistent with the officially published calibration coefficients， with an average relative difference of less than 2.25%. The calibration uncertainty is within 5.34%. 2） The long time series cross calibration results showed that after one year on-orbit operation， the gain status of HJ-1A/CCD1 was adjusted on October 20， 2009， resulting in a sudden change in the cross-calibration coefficients on the 409th day after launch. After the adjustment， the overall radiometric performance was relatively stable. 3） The radiometric performance of HJ-1A/CCD1 showed a slow and fluctuating downward trend from October 2009 to December 2021， with an annual attenuation rate of less than 3.10%. The method proposed in this paper can effectively improve the radiometric calibration frequency and accuracy， and can be used for radiometric performance monitoring over the whole life cycle of HJ-1A/CCD1.

Machining feature recognition is crucial in computer-aided design （CAD） and manufacturing （CAM） as it serves as a vital link between CAD and CAM systems. Researchers have proposed two types of machining feature recognition methods： rule-based and learning-based， with the latter showing superior performance and garnering more attention. However， existing recognition methods face challenges such as insufficient utilization of geometric information， inaccurate machining feature localization， and complexity in the instance segmentation process. To address these issues， this paper proposes PT-MFR， a CAD model machining feature recognition method based on Point Transformer. It performs two tasks： semantic segmentation and instance segmentation， predicting the semantic category of machining features for each face and calculating face similarity to segment the machining feature instances. The results of both tasks are integrated to obtain the machining feature recognition results. Experimental results demonstrate that the proposed method outperforms other methods.

The detection of maneuvering weak targets by monostatic ground-based radar is a hot issue， and the detection performance of moving targets can be improved by increasing the accumulation time. However， with the increase of accumulation time， the movement of maneuvering target causes the phenomenon of range cell migration （RCM） and Doppler frequency migration （DFM）， which leads to the loss of accumulation gain. To solve this problem， a long-time accumulation method for weak maneuvering target detection in monostatic radar is proposed in this paper. The segmentation strategy is firstly given in the proposed method， then the motion compensation algorithm with the second-order keystone transform is used twice to correct the RCM， later LVD is used to estimate the motion parameters and the phase compensation is carried out. Finally， the detection probability and detection accuracy are further improved by sliding window non-coherent accumulation. The simulation results show that the detection performance of the proposed method is better than the traditional accumulation detection algorithm in the weak maneuvering target condition. The proposed method compensates for the gain loss caused by RCM and DFM and improves the signal-to-noise ratio of coherent and non-coherent accumulation gain. Furthermore， it can balance the computational complexity and accumulation gain at the same time.

21 centimeter array （21CMA） is China’s pilot equipment in the square kilometre array （SKA） low-frequency band， therefore， the National SKA Program of China launches a special upgrade plan for 21CMA to enable pulsar observation capability. To solve the problem of receiving and storing massive observation data after upgrading， the project designed and introduced a distributed file storage system based on advanced RISC machine （ARM）， but there are still shortcomings in terms of usability and standardization. This study customizes deployment， failure recovery， and system monitoring interface module for 21CMA storage device， and carries out system tests. The test covers the deployment time， failure recovery time， and database stability of the monitoring system. The test results show that the system solves the operation and maintenance management problems of 21CMA storage devices， also improves their reliability and efficiency， meets the needs of 21CMA’s multi-cluster monitoring， and is of great significance for similar projects in the future.