The rise of vehicle-infrastructure (V2I) collaboration enables safer and broader perception. To process large-scale V2I video streams, vision-language models (VLMs) are promising as they unify multi-view vision into end-to-end task grounding, reducing handcrafted design. We use Vision Mixture-of-Experts (V-MoE) as the distributed visual backbone of VLMs, leveraging sparse expert routing to enable conditional computation across diverse viewpoints under resource constraints. Yet, V-MoEs face a critical challenge: large-scale data shifts over minutes to hours in V2I systems, amplified by agnostic participants and biased features propagating through experts. To maintain accuracy efficiently, we find it beneficial to co-adapt multiple V-MoEs on edge servers, avoiding the latency and privacy risks of cloud offloading and the accuracy sacrifices of on-device methods. However, the resource-constrained edge poses challenges for efficient co-adaptation: i) DRAM fragmentation and imbalance limit expert parallelism, ii) memory-I/O bottlenecks restrict computation reuse, and iii) asynchronous adaptation increases task-switch overhead. Also, prior work rarely explores the upper bound of concurrent tasks under limited edge resources, a critical factor for practical V2I deployment. To address these, we present AdaSprite. By combining cooperative elastic scaling with multi-level multiplexing, AdaSprite optimizes expert lifespans to reduce DRAM fragmentation, exploits predictable activation patterns for efficient I/O reuse, and employs twin-buffer scheduling to leverage sparsity. On a weak edge, AdaSprite supports up to 17 concurrent V2I tasks (vs. up to 6 for baselines), improving SLO attainment by 1.6x and throughput by 2.1x. Also, it allows users to trade accuracy and concurrency for second-level adaptation.

The rise of mobile devices equipped with numerous sensors, such as LiDAR and cameras, has driven the adoption of multi-modal deep intelligence for distributed sensing tasks, such as smart cabins and driving assistance. However, the arrival time of mobile sensory data vary due to modality size and network dynamics, which can lead to delays (if waiting for slow data) or accuracy decline (if inference proceeds without waiting). Moreover, the diversity and dynamic nature of mobile systems exacerbate this challenge. In response, we present a shift toopportunisticinference for asynchronous distributed multi-modal data, enabling inference as soon as partial data arrives. While existing methods focus on optimizing modality consistency and complementarity, known as modal affinity, they lack acomputationalapproach to control this affinity in open-world mobile environments.${\sf AdaFlowLite}$pioneers the formulation of structured cross-modality affinity in mobile contexts using a hierarchical analysis-based normalized matrix. This approach accommodates the diversity and dynamics of modalities, generalizing across different types and numbers of inputs. Employing an multi-modal lightweight Swin Transformer (MMLST),${\sf AdaFlowLite}$facilitates real-time and flexible data imputation, adapting to various modalities and downstream tasks without retraining. Experiments show that${\sf AdaFlowLite}$significantly reduces inference latency by up to 80.4% and enhances accuracy by up to 62.1%, while achieving nearly a 50% reduction in energy consumption, outperforming status quo approaches. Also, this method can enhance LLM performance to preprocess asynchronous data.

The automatic intelligent identification of open-pit mines could improve the efficiency in the supervision of mines. In this paper, a massive sample database of open-pit mines was firstly established with remote sensing resolution of 0.5 m, 2 m from 2018 to 2021. Then, an intelligent identification algorithm for damaged land of open-pit mines was constructed with space pyramid, cavity convolution and other structures considered in the design part. Taking Zhongliang mountain of Chongqing as a study area, the manual visual interpretation results were applied to validate the cavity convolution model. The result showed that total precision of algorithm was more than 85% which improved the identification and extraction ability of open-pit mines. The precision with remote sensing image resolution of 0.5m were higher than that of 2m. The efficiency was 18 times compared with manual visual interpretation.

In this study, the massive tilt data loading technique based on multi-parameter rendering is deeply researched and explored. First, to address the needs and challenges of tilt data loading, this paper clarifies the purpose and significance of the study based on the research background. Then, by analyzing the characteristics and needs of tilted data loading, a data loading technique is proposed based on the control of multiple parameters such as real-time network bandwidth and camera view angle height, and the method and process of this technique are described in detail. Then, this study experiments and analyzes the newly proposed loading technique and draws encouraging results and findings. Finally, this paper summarizes the conclusions drawn from the study and emphasizes the importance and potential applications of this research for massive inclined data processing. Through this paper, we expect to provide a more efficient and accurate tilt data loading technique for the fields of digital twin, urban planning and 3D geographic information.

A gradual refinement strategy method to build a large-scale landcover semantic segmentation dataset of remote sensing image based on the existing monitoring data and historical orthophotos was proposed in this paper. Firstly, a fixed-size sliding window was used to slide on the remote sensing image and the coverage mask obtained from the monitoring vector data at the same time to extract the image slice and label slice, and the original dataset was composed of a large number of slices. Secondly, the original dataset was divided into several subsets by DBSCAN, and the error samples covered by clouds or shadows were effectively classified and deleted from the dataset. Then, the low-precision samples were detected and deleted from the dataset by an optimization and purification method based on iterative optimization intelligent model. Experimental results show that the method proposed in this paper could build a large-scale remote sensing image semantic segmentation dataset effectively based on historical orthophoto images and the monitoring vectors which were not strictly matching the images.

The study on the influence of urban building forms on land surface temperature (LST) can provide a scientific reference for building distribution optimization in mountain cities. This paper obtained the urban building properties such as building types, building floors from field investigation based on high-resolution remote sensing images. The LST was calculated by the Radiative Transfer Equation method based on Landsat-8 data in 2021. We constructed five factors of urban building forms (average building density, average building height, floor area ratio, average building volume, and sky view factor) to analyze the correlation between urban building forms and LST. The result showed, only average building density had a certain correlation with LST, while other factors had a small correlation which indicated some factors of urban building forms may not fully reflect the characteristics of LST in mountain cities. By constructing multiple linear regression models between five factors of urban building forms and LST, we found that when the number of factors increased, the more accurate the models were, with higher the computational complexity. However, the increase of model accuracy gradually slowed down when the number of factors exceeded 3. It was appropriate to choose linear regression models of 2-factors or 3-factors to simulate LST.

The rise of sensor-rich mobile devices has led to the adoption of multi-modal deep intelligence for distributed sensing tasks. However, varying arrival times of mobile sensory data can cause delays or accuracy decline. The diversity and dynamic nature of mobile systems further exacerbate this challenge. To address this, we present an opportunistic inference approach for asynchronous distributed multi-modal data, enabling inference upon partial data arrival. AdaFlow pioneers the formulation of structured cross-modality affinity using a hierarchical analysis-based normalized matrix, accommodating the diversity and dynamics of modalities. Employing an affinity attention-based conditional GAN (ACGAN), AdaFlow facilitates flexible data imputation, adapting to various modalities and tasks without retraining. Experiments show significant reductions in inference latency and enhanced accuracy compared to status quo approaches.

The rise of mobile devices equipped with numerous sensors, such as LiDAR and cameras, has spurred the adoption of multi-modal deep intelligence for distributed sensing tasks, such as smart cabins and driving assistance. However, the arrival times of mobile sensory data vary due to modality size and network dynamics, which can lead to delays (if waiting for slower data) or accuracy decline (if inference proceeds without waiting). Moreover, the diversity and dynamic nature of mobile systems exacerbate this challenge. In response, we present a shift to opportunistic inference for asynchronous distributed multi-modal data, enabling inference as soon as partial data arrives. While existing methods focus on optimizing modality consistency and complementarity, known as modal affinity, they lack a computational approach to control this affinity in open-world mobile environments. AdaFlow pioneers the formulation of structured cross-modality affinity in mobile contexts using a hierarchical analysis-based normalized matrix. This approach accommodates the diversity and dynamics of modalities, generalizing across different types and numbers of inputs. Employing an affinity attention-based conditional GAN (ACGAN), AdaFlow facilitates flexible data imputation, adapting to various modalities and downstream tasks without retraining. Experiments show that AdaFlow significantly reduces inference latency by up to 79.9% and enhances accuracy by up to 61.9%, outperforming status quo approaches. Also, this method can enhance LLM performance to preprocess asynchronous data.

The monitoring of vegetation recovery was an important factor to evaluate the effectiveness of ecological restoration of open-pit mines. Lidar data can obtain the horizontal and vertical structures of trees that was wildly used to characterize the 3-dimensional (3D) structure of trees. In this paper, taking Ba’nan District of Chongqing as a study area, high density lidar data was applied to obtain the vegetation vertical structure parameters based on marker-controlled watershed segmentation method with IPTD filtering algorithm. The result showed, average tree height, diameter at breast height (DBH), crown diameter of open-pit mines was relatively lower compared with other areas outside as the trees of open-pit mines were mainly artificial planting and the growth time of trees was about 3-5 years. The trees had recovered well in the process of restoration. It indicated that the ecological restoration and planting of trees in open-pit mines were still in the process of recovery.

Constructing a scientific evaluation system of geological environment and restoration can provide a reliable theoretical basis for mine geological environment protection. In this paper, we firstly constructed the evaluation system of 10 indicators and 30 sub-indicators as the index of ecological environment and restoration. The hierarchical analysis method was used to evaluate each mine according to the actual situation in Chongqing. We took Mengte mine as a study area and obtained the parameters of soil, water, vegetation, and biology by field experiment. Analytic hierarchy process was applied to this paper and the weight of each indicator was calculated based on expert scoring method and the experience method. This research provided technical support for geological environmental assessment in other mines.

At present, there are two methods to calculate the microwave extinction cross-section, one is based on the energy conservation which could be calculated from backscattering cross-section, another is based on forward scattering theory in accordance with the imaginary part of the scattering amplitude in the forward direction. In this study, the extinction cross-section of two methods are employed to obtain single scattering albedo and optical depth. The 0th order Radiative Transfer Model which considers the vegetation layer as a uniform medium is used to calculate the brightness temperature of the vegetation covered ground. The simulating results are validated by observation data based on the experiments of typical vegetation (cotton and soybean). The result shows, the simulations of both methods matches the field measurements well, but the forward scattering theory seems more accurate. However, when the incident angle becomes bigger, the results of two methods tend to be very different, the energy conservation method is lower than the observation while the forward scattering method is higher.

A new method was proposed in this paper to estimate optical depth (τ) and single scattering albedo (ω) of short vegetation over north China plain based on the refined physical expressions of Microwave Vegetation Indices derived from the parameterized first-order emission model. Comparisons with MODIS 16-day Normalized Difference Vegetation Index (NDVI) showed that, although there were some differences between the variations of single scattering albedo, optical depth and NDVI, the variation trends of the three parameters are very similar to each other. They all mainly expressed two regular variations during the whole year of 2010 that firstly increased and then decreased. The first one happened during February to late June, and the second one happened during early July to late September. They are closely consisted with the phenology of winter wheat and summer corn, which are the mainly two crop types in north China plain.

Dielectric constant is an important parameter in microwave remote sensing. The microwave scattering/radiation signal of vegetation is closely related to its dielectric constant. Many related models have been established by now. However, most of them can only be used in room temperature. Therefore, it brings errors in the research of vegetation in frozen environment. In this study, a new dielectric model, which can be used at frequencies ranged from 3GHz to 40GHz under temperatures between -20°C and -4°C, has been established. It was developed based on Debye-Cole dual-dispersion model. The validation shows it has an acceptable precision.

In this study, the dielectric constants of 12 types of soil with different organic matter content were measured using the coaxial probe method by network analyzer (0.5-40 GHz) at room temperature (approx. 23°C). The observed dielectric constant increases only slowly with soil volumetric water content up to a transition point. Beyond the transition point, it increases rapidly with volumetric water content. It was found that the value of the transition point was higher and the observed dielectric constant was lower at the same soil volumetric water content and frequency for soil with higher organic matter content. A simple semi-empirical model was proposed to describe the dielectric behavior of soil with organic matter. This model was developed based on the refractive mixing dielectric model (RMDM).

A new dielectric model for vegetation in frozen environment based on the Debye-Cole dual-dispersion model was already developed in part I. This model can be used at a wide frequency range (0.5GHz - 40GHz) and even applicable for negative temperatures reached -20°C. In this paper, a matrix-doubling microwave emission model was used to evaluate vegetation effects in a frozen environment at 6.925, 10.65, 18.7 and 36.5GHz (V and H polarization). To verify the new developed dielectric model, a kind of young tree named Populus tomentosas was measured based on the Truck-mounted Multi-frequency Microwave Radiometer in December of 2012. In the experiment, the ground was irrigated to get rid of the soil signals. Also, the row-structure's effect on the trees can be eliminated when water covered the whole ground surface. Comparisons and analysis between model simulations and field measurements showed the dielectric model can be applied to microwave emission model as input data. Furthermore, the characteristics of microwave radiation of vegetation in frozen environment were evaluated and how the vegetation dielectric constant affected the electric field and physical property of vegetation layer was explained.

Snow cover information is crucial to global climate change research and hydrological applications. Snow cover over the Tibetan Plateau is important to water resources and Asian climate. Based on high temporal resolution of geostationary satellite data, snow cover map with less cloud obscuration can be obtained daily. In this paper, geostationary meteorological satellite FY2E VISSR data is used to obtain the snow cover information over the Tibetan Plateau in year 2010 and 2011 winter seasons. Meteorological station observations are used to evaluate the performance of snow cover maps. In addition, MODIS and IMS snow cover products are used for comparison and validation. Results indicate VISSR snow cover maps show good performance in reducing cloud obscuration. MODIS snow cover maps present highest overall accuracy, followed by VISSR and IMS. VISSR and IMS snow cover maps show slight over-estimation of snow cover over the Tibetan Plateau.

In this study, a parameterized first-order radiative transfer (RT) model for short vegetation layer is employed to retrieve the single scattering albedo of winter wheat by combining passive microwave AMSR-E data with optical MODIS data. The microwave vegetation indices (MVIs) with two adjacent frequencies of AMSR-E at H/V polarization derived from the parameterized model are used to cancel out the ground surface emission signals. Then a simulating database based on field measured parameters is established to figure out the relationship of the optical thickness, single scattering albedo at C and X band, respectively. Finally the characteristics of retrieved single scattering albedo are analyzed and the daily NDVI was utilized for evaluating the retrieved results.

Surface soil moisture is a key parameter in material exchange and energy cycle at the land surface and atmosphere interface. In this study, a soil moisture retrieval algorithm was developed based on a parameterized first-order radiative transfer (RT) model. This method was validated in ground surface covered by vegetation in Tibet. The results showed that the RMSE of the retrieved soil moisture reached about 4% by using the parameterized first-order RT model. The retrieval results were also compared to zero-order RT model and noted that both zero-order and first-order RT model varied similarily but the first-order retrieval algorithm seemed to be more accurate than zero-order model. Future work will investigate possible improments to the algorithm extend testing of the algorithm to other regions.

Simulation of brightness temperature and related snow parameters is essential to understand the microwave emission property and its evolution with change of the snow soil system status. In this paper, a typical thin snow pack on North China Plain is measured on Nov 13-16th, 2009 at Luancheng test site HUT (Helsinki University of Technology) wet snow emission model is used to predict the brightness temperatures at 10.65, 18.7 and 36.5 GHz. A physically-based snow process model, SNTHERM (SNow THERmal Model), is applied to simulate the snow melting process. The measured snow density and grain size is compared with SNTHERM prediction and HUT inputs. Results show that the application of snow emission model and process model can explain the variation trend of wet snow emission properties well.