We propose a semi-supervised semantic segmentation framework based on a novel self-training method with Ambiguity-Driven Online Refinement (ADOR). Compared to recent methods based on consistency regularization that only use a limited amount of unlabeled data due to the need to filter noisy pseudo-labels, our approach leverages a larger amount of unlabeled data for training while being able to dynamically refine pseudo-labels to mitigate noise. To achieve this, we introduce Spatial Ambiguity Resolution (SAR) that selectively filters and refines the ambiguous regions in pseudo-labels produced by CNNs and ViTs. The proposed method leverages both local and global confidence as a robust estimate for label quality, and a stochastic correction disrupts the accumulation of errors. In addition, we present Semantic Ambiguity Learning (SAL) to adaptively enhance class separability based on varying levels of pairwise class confusion. Specifically, we adjust the pairwise similarity between a query feature and its positive and negative class prototypes, in order to place more emphasis on classes that are prone to confusion. Experimental results on both the Pascal VOC and Cityscapes datasets demonstrate the efficacy of our method with both CNNs and ViTs. Code will be made publicly available at https://github.com/LlistenL/ADOR.
Abstract Green spaces in densely urbanized areas along China’s southeast coast generally face spatial competition between ecological protection and development intensity, as well as conflicts between policy-oriented and market-driven values. Among them, Changle District in Fuzhou City, as the core area of the coastal new city, needs to coordinate its green space system to address complex three-dimensional problems. To explore the optimal configuration of green spaces under different ecological constraints and achieve a balance between ecological, economic, and social benefits, this paper takes Changle District in Fuzhou City as the research area, with 2020 as the base year, and innovatively combines a multiobjective optimization model with the Geographic Simulation and Optimization System–Future Land Use Simulation (GeoSOS-FLUS) model. The former is used to balance multiple objectives such as ecological conservation, economic development, and social needs, generating optimal allocation schemes. The latter precisely simulates the dynamic evolution of green spaces based on these schemes. Together, they overcome the limitations of traditional models, which cannot account for economic and ecological benefits when simulating complex urban land use. Based on this model combination, this paper simulates and predicts the area structure, spatial configuration, and target benefits of green space in Changle District in 2030 under ecological high-, medium-, and low-constraint scenarios. The results reveal the following: (1) In terms of area structure, the area of cropland and unused land in the three scenarios is smaller than the current value in 2020, while forest land and grassland have increased, water has been effectively protected, and construction land has gradually increased as ecological constraints have been relaxed. (2) In terms of spatial distribution, the area of cropland in the north, west, and central parts of the study area decreased significantly in all three scenarios. At the same time, the increase in forest land and grassland was concentrated in the north and west, and the rise in water was focused on the central part of the study area. The area of construction land in the south and west decreased slightly under the ecological high- and medium-constraint scenarios. At the same time, it increased around the center of the study area under the low ecological constraint scenario. (3) From the perspective of target benefits, under the high-ecological-constraint scenario, the ecological benefits and economic benefits are 4.122 and 94.545 billion yuan, respectively, which is a significant increase compared with the ecological benefits in 2020, while the economic benefits have slightly decreased. Under the medium-ecological-constraint scenario, the ecological and economic benefits are 4.067 and 104.023 billion yuan, respectively, representing significant increases compared with 2020. Under the low-ecological-constraint scenario, the ecological benefits and economic benefits are 4.009 and 112.945 billion yuan, respectively, which represents a slight increase in ecological benefits compared to 2020 but a significant increase in economic benefits. The conclusions of this study can be applied to urban green space planning practices, providing comprehensive support for decision-makers in formulating scientific and reasonable green space allocation strategies under different ecological constraints and helping to achieve sustainable urban development goals.
Although recent research has documented that disclosure of pollution information shapes human avoidance behavior, there is limited knowledge on how disclosure of pollution information influences firms’ environmental performance. Taking the release of the Chinese documentary “Under the Dome” in 2015 as an exogenous quasi-natural experiment and adopting a difference-in-differences (DID) model, this paper investigates whether and how disclosure of pollution information shapes firms’ environmental performance. Results show that disclosure of pollution information strengthens firms’ environmental disclosure but does not lead to better environmental practices, thereby providing clear evidence of ESG decoupling. The potential channels include raising environmental awareness among firms’ external stakeholders by increasing media coverage, heightening analyst following, and strengthening environmental regulation. Additionally, the effect is particularly prominent for firms with high reputation, low market pressure, low managerial career concerns, and for firms in labor-intensive industries. Overall, this paper highlights the pivotal role that disclosure of pollution information plays in improvingenvironmental governance.
Long-range (LR) interactions are no longer just a theoretical idea: they can be engineered in several low-dimensional platforms and offer a new way to control heat transport. At the same time, they push beyond the standard picture developed mainly for short-range, momentum-conserving lattices. In this review, we consider one-dimensional Fermi-Pasta-Ulam-Tsingou (FPUT)-type lattices where LR effects enter mainly through a quartic anharmonic coupling that decays as a power law, characterized by an exponent s. Our goal is to explain, in a unified way, how LR anharmonicity changes microscopic energy exchange, collective dynamics, and the macroscopic scaling laws of heat conduction-while also clarifying what is well established and what is still debated. We first introduce the main models and the quantities used to discribe transport in LR-FPUT systems. These include the nonequilibrium steady-state thermal conductivity i(N), the equilibrium Green-Kubo (GK) approach based on the heat-current autocorrelation CJJ(t), and spatiotemporal correlations of heat/energy fluctuations pQ(m, t) , together with the scaling of the mean-squared displacement (MSD). A key technical point is that, in LR lattices, the microscopic heat current contains explicit nonlocal contributions. As a result, different (formally related) implementations of the current may lead to noticeable differences at finite system sizes and finite times. To extract the intrinsic transport behavior, we therefore stress practical checks: i) the stability of scaling exponents against boundary and driving protocols, ii) validation among all the three methods, i.e., i(N) , CJJ (t) , and pQ(m, t) , and iii) careful control of finite-size and finite-time crossovers. Based on these methods, we review representative s-dependent transport regimes. i) The special point a = 2 is central and remains controversial. Early work reported ballistic-like behavior under certain boundary conditions. More recent studies using more controlled numerical method (e.g., periodic driving together with correlation scaling) tend to support strongly enhanced superdiffusion, with an effective divergence exponent around alpha similar or equal to 0.7 , rather than ballistic transport. The most consistent physical picture links this enhancement to weak nonintegrability and the presence of long-lived coherent carriers, such as mobile breathers or related nonlinear excitations, while recognizing that finite-size results are unusually sensitive to boundary implementation and to how the heat current is defined. We summarize the current consensus and the main remaining disagreements, and we argue that ballistic vs strong superdiffusion should be decided by consistency across multiple observables, not by temperature profiles alone. ii) In the weak-LR regime (1 <=sigma <= 3), a striking result is a near-diffusive window around a similar or equal to 1.25. This is conceptually important because it suggests a possible route toward quasi-normal transport within Hamiltonian, momentum-conserving one-dimensional lattices, which usually show anomalous conduction in the short-range case. We discuss how this challenges the common intuition that momentum conservation implies anomalous transport, but we also emphasize that the current evidence should be viewed as a strong candidate that still needs careful confirmation against finite-size crossovers and method dependence. iii) In the strong-LR regime (0 <= sigma <= 1), simulations show clear antipersistent (negative) structures in CJJ(t) and a trend toward slower transport, sometimes described as subdiffusive-like transport. iv) We also discuss an inverse LR coupling design LR-FPUT model that can invoke ballistic transport, suggesting that LR nonlocality combined with structural modulation may open new transport channels and may be useful for thermal rectification. Finally, we outline key open directions: establishing reliable finite-size scaling to distinguish genuine sdriven dynamical transitions from broad crossovers; mapping transport regimes in a (sigma, T, nonlinearity) phase diagram; and identifying the microscopic carriers (phonons, stationary/mobile breathers) and their scattering dynamics that underpin the observed superdiffusive, near-diffusive, and antipersistent slow-transport behaviors. These needs also connect to experiments, since LR couplings can be realized in platforms such as Coulomb crystals and frustrated magnets, and LR interaction engineering has already enabled efficient thermal rectification-suggesting practical routes to tune heat conduction and build thermal diodes/switches based on nonlocality and nonlinearity.
Electron donor-acceptor (D-A) complexes are emerging as promising candidates for the development of effective nonlinear optical limiting (OL) materials but their preparation still remains challenging. Herein, we report an electron D-A OL device based on a metal-porphyrinic framework thin film loaded with carbon nanodots (CDs). The composite film was prepared by impregnating a porous PIZA-1 MOF film with an o-phenylenediamine precursor and subsequent in situ conversion of the precursor into CDs via laser irradiation. The resulting o-CDs@PIZA-1 thin film exhibits superior OL performance with a lower OL threshold of 2.21 J cm-2 and a higher nonlinear absorption coefficient of 3.1 & times; 10-6 m W-1 compared to the pristine PIZA-1 thin film. This enhancement is mainly attributed to energy transfer/photoinduced electron transfer (ET/PET) of the porphyrin group and the confined CDs. Density functional theory (DFT) calculation reveals significant electronic interactions between the PIZA-1 moiety and CDs, with electron transfer pathways facilitating charge transport and optimizing the overall performance. This study proposes a novel MOF-nanocomposite strategy that effectively incorporates CDs into a solid matrix, enabling the preparation of high-performance OL materials ready for device integration and offering a viable route toward practical applications.
