The arms race between Internet censorship and evasion has pushed censors from static rule-based filtering to deep learning-based traffic analysis. Recent automated evasion tools still suffer from insufficient robustness across diverse censorship modalities and poor usability, requiring manual fitness tuning or domain-specific languages. We reframe censorship evasion as a semantic image-to-image editing task executable with a single prompt, and introduce FlowPaint, a generative framework that leverages the "world knowledge" of large diffusion models to reshape malicious traffic into benign patterns.

We present a quantum-inspired algorithm that utilizes Quantum Hamiltonian Descent (QHD) for efficient community detection. Our approach reformulates the community detection task as a Quadratic Unconstrained Binary Optimization (QUBO) problem, and QHD is deployed to identify optimal community structures. We implement a multi-level algorithm that iteratively refines community assignments by alternating between QUBO problem setup and QHD-based optimization. Benchmarking shows our method achieves up to 5.49% better modularity scores while requiring less computational time compared to classical optimization approaches. This work demonstrates the potential of hybrid quantum-inspired solutions for advancing community detection in largescale graph data.

Distributed quantum computing (DQC) that allows a large quantum circuit to be executed simultaneously on multiple quantum processing units (QPUs) becomes a promising approach to increase the scalability of quantum computing. It is natural to envision the near-future DQC platform as a multi-tenant cluster of QPUs, called a Quantum Cloud. However, no existing DQC work has addressed the two key problems of running DQC in a multi-tenant quantum cloud: placing multiple quantum circuits to QPUs and scheduling network resources to complete these jobs. This work is the first attempt to design a circuit placement and resource scheduling framework for a multi-tenant environment. The proposed framework is called CloudQC, which includes two main functional components, circuit placement and network scheduler, with the objectives of optimizing both quantum network cost and quantum computing time. Experimental results with real quantum circuit workloads show that CloudQC significantly reduces the average job completion time compared to existing DQC placement algorithms for both single-circuit and multi-circuit DQC. We envision this work will motivate more future work on network-aware quantum cloud.

When given two similar images, humans identify their differences by comparing the appearance (e.g., color, texture) with the help of semantics (e.g., objects, relations). However, mainstream binary change detection models adopt a supervised training paradigm, where the annotated binary change map is the main constraint. Thus, such methods primarily emphasize difference-aware features between bi-temporal images, and the semantic understanding of changed landscapes is undermined, resulting in limited accuracy in the face of noise and illumination variations. To this end, this paper explores incorporating semantic priors from visual foundation models to improve the ability to detect changes. Firstly, we propose a Semantic-Aware Change Detection network (SA-CDNet), which transfers the knowledge of visual foundation models (i.e., FastSAM) to change detection. Inspired by the human visual paradigm, a novel dual-stream feature decoder is derived to distinguish changes by combining semantic-aware features and difference-aware features. Secondly, we explore a single-temporal pre-training strategy for better adaptation of visual foundation models. With pseudo-change data constructed from single-temporal segmentation datasets, we employ an extra branch of proxy semantic segmentation task for pre-training. We explore various settings like dataset combinations and landscape types, thus providing valuable insights. Experimental results on five challenging benchmarks demonstrate the superiority of our method over the existing state-of- the-art methods. The code is available at SA-CD.

Key-value lookup functions have been widely applied to network applications, including FIBs, load balancers, and content distributions. Two key performance requirements of a lookup algorithm are high throughput and small memory cost. One limitation of existing fast network lookup algorithms is that they require multiple independent and uniform hash functions, which cost high computation time and might not be available on existing hardware network devices. Recently developed learned model hashing (LMH) proposes to use a linear machine learning model to replace hash functions to avoid hash computation, but they are not optimized for memory cost. We propose a novel network lookup method called Parrot hashing, which uses a learned model to distribute keys into different buckets and applies a simple perfect hashing method to resolve the collisions of the keys in a bucket. Parrot can be implemented with only one CRC-8, which is available on all network devices. We implement Parrot in three prototypes: a software program on end hosts, a software switch, and a FIB running on a hardware programmable switch. The experimental results show that Parrot achieves the highest lookup throughput on all three prototypes, compared to existing methods. Its memory cost is also significantly lower than that of LMH.

LSM tree based stores are a popular database design in modern persistent storage systems due to their efficient writes with sorted keys. However, this hierarchical log structure suffers from extensive read amplification because multiple disk accesses are required when it searches for a key. Recent optimizations of LSM trees propose caching hot keys to reduce I/Os mainly based on their access frequencies. However, our empirical studies show that keys are different in I/O costs, which should also be considered in the caching policy: caching key-value pairs with high I/O cost can effectively improve query latency. In addition, false positives incurred by the Bloom filters in LSM trees introduce a large overhead to access SSTables because the queried keys do not exist. In this work, we design and implement SpotKV, which resolves the above two problems in an LSM tree store by proposing two memory-efficient data structures, weighted Count- Min sketch for access and I/O-aware cache admission and dynamic-seed Cuckoo filters for eliminating false positives, to improve data lookup throughput. We implement SpotKV on Google's LevelDB vl.20. From extensive experimental evaluations, SpotKV achieves 1.2-3.0x read throughput while using the same or smaller memory, compared with several state- of-the-art LSM tree stores under the read-heavy workloads of the YCSB benchmarks.

Key-value lookup functions have been widely applied to network applications, including FIBs, load balancers, and content distributions. Two key performance requirements of a lookup algorithm are high throughput and small memory cost. One limitation of existing fast network lookup algorithms is that they require multiple independent and uniform hash functions, which cost high computation time and might not be available on existing hardware network devices. Recently developed learned model hashing (LMH) proposes to use a linear machine learning model to replace hash functions to avoid hash computation, but they are not optimized for memory cost. We propose a novel network lookup method called Parrot hashing, which uses a learned model to distribute keys into different buckets and applies a simple perfect hashing method to resolve the collisions of the keys in a bucket. Parrot can be implemented with only one CRC8, which is available on all network devices. We implement Parrot in three prototypes: a software program on end hosts, a software switch, and a FIB running on a hardware programmable switch. The experimental results show that Parrot achieves the highest lookup throughput on all three prototypes, compared to existing methods. Its memory cost is also significantly lower than that of LMH.

In order to give full play to the application efficiency of land remote sensing satellites for natural resources and strengthen the protection of cultivated land in China, this paper introduces a two-stage intelligent extraction strategy for erosion gullies. This strategy integrates deep learning object detection with interactive semantic segmentation.Initially, an iterative optimization method of object detection samples and models based on incremental learning and confidence filtering is proposed, achieving automatic identification of erosion gullies. Then, an interactive semantic segmentation method based on edge constraint is adopted to extract the precise contours of erosion gullies semi-automatically. An erosion gully extraction experiment was conducted in Heilongjiang Province of China, utilizing satellite images with a spatial resolution of 2 meters. Accuracy assessment was performed using GF-7 satellite imagery and verified through field verification. The results indicate that the Precision of erosion gully extraction in test area was 95.4%, the Recall was 93.0%, and the F1_score was 94.2%. The feasibility and accuracy of the proposed method has been verified by the experiments.

As an important part of the renewable energy, photovoltaic power generation industry has developed rapidly all around China in recent years, however some land use problems have also emerged. Therefore it is of great significance to monitor the number and distribution of photovoltaic power stations timely and accurately with high-resolution satellite images for the healthy development of photovoltaic industry. Combined with the improved DeepLab V3+ model and the ResNeSt-50 backbone network, the paper designs an effective photovoltaics extraction semantic segmentation algorithm and trains the new extraction model iteratively by making full use of big and various photovoltaic land samples. Photovoltaics are extracted accurately all over China with Chinese high-resolution satellite images, following a series of post-processing algorithms, such as binarizing, small and pseudo targets automatic removing, etc. Results show that the accuracy rate of photovoltaic land extraction is about 72.36% and the recall rate is about 91.06%. This precision is good enough for photovoltaic land extraction nationwide annually and the proposed deep learning model is efficient, small and can widely be used with other natural resources target extraction.

Surface water is an irreplaceable strategic resource for human survival and social development. It is of great significance to fully grasp the quantity, spatial distribution and dynamic changes of surface water in China accurately, quickly and timely. With the rapid development of Chinese remote sensing satellite industry, monitoring surface water in whole China quarterly has become a task of challenge but achievable. This paper analyzes and compares the advantages and disadvantages of existing algorithms for surface water extraction. And a practical and useful algorithm workflow is proposed. Based on 2-meter resolution multi-spectral satellite images, the paper adopts the regional fast-growing water extraction algorithm combined with water element buffers. 1,265 above level 3 rivers, their associated 2,128 reservoirs and 2,953 above-1-km2 natural lakes in China have been fast extracted quarterly with high-precision. Practice has shown that the automatic extraction algorithm is one of the most effective algorithms for realizing the automatic monitoring of large-scale surface water.

At present, the Synthetic Aperture Radar (SAR) is one of the few remote sensing methods that can realize rapid all-day and all-weather mapping in difficult areas of surveying and mapping, and has unique advantages incomparable with traditional optical remote sensing technology. And SAR polarimetric scattering mechanism provides important theoretical basis for surface radar target recognition. The paper adopts the P-band polarization data acquired by Chinese Academy of Surveying and Mapping SAR system (CASMSAR) as well as the C-band polarization data from Radarsat-2 and use the Freeman-Durden decomposition method to analyze and compare the difference of six typical terrain targets' scattering characteristics of different band SAR images. Results are as follows: (1) the volume scattering values of vegetation at different height are relatively high in C-band SAR image, which can be distinguished from non-vegetation, but the scattering characteristics between vegetations are confusing. (2) For P-band SAR image, there is a great difference in the volume scattering characteristics between vegetations at different height, while little difference in the odd scattering characteristics among bare soil, water area, and burnt wheat fields. (3) The volume scattering characteristics of vegetations at high height are obvious in P-band SAR image, while those at low height are obvious in C-band SAR image.

Remote sensing derived water area and volume have been widely used in large lakes monitoring. The seed point method was used to monitor the water area and volume of Poyang Lake, as an important international wetland and the largest fresh water lake in China, from 2019 to the flood season in 2020 using ZY3, GF-1, GF-3, GF-6, and BJ-2 data. Furthermore, historical water area and water volume changes of Poyang Lake were analyzed utilizing public data set. The results showed that the water area and water volume of Poyang Lake changed significantly in recent sixty years. It also fluctuated violently during the high water and low water seasons from 2019 to 2020. In the past sixty years, the minimum and maximum water area was 1190.73 km2in 1960 and 31 79.31 km2in 2019, respectively, and the water volume increased by 1853.62 million cubic meters. The water area difference of Poyang Lake between the high water season and the low water season was more than quadruple, which varied from 3179.21 km2in the third quarter in 2019 to 674.01 km2in the first quarter in 2020, and the water volume decreased by 489.95 million cubic meters. The flood in 2020 inundated 7834.15 hectares of planting land, 2300.81 hectares of forest and grass coverage, 906.51 hectares of desert and bare land, 2170.87 hectares of water area, 101.78 hectares of housing construction area, and 73.15 hectares of railway and highway. The relevant results provide a scientific basis for the construction of Poyang Lake ecological economic zone and the ecological protection of Poyang Lake wetland and a decision-making reference for flood control and disaster relief in Poyang Lake area.

Mangroves provide a variety of irreplaceable functions such as coastline protection, bay improvement, water purification and wetland diversity protection. However, mangroves are fragile ecosystems and are affected by human activities and climate change. Long-term monitoring on mangroves is of great significance. This paper selected Tongming Bay located in southern China coast as a study case because of its high variation of mangrove extent. Satellite remote sensing images from 1978 to 2018 were used for mangrove extent interpretation, and to obtain the distribution, changes, and main driving factors of changes in the Tongming Bay area in the past 40 years. The results showed that the area of mangrove extent in Tongming Bay continued to shrink from 1978 to 2013, and the area gradually increased after 2013. The main driving factor for the reduction of mangroves was the artificial aquaculture occupation of mangrove habitats, and mangrove area growth was mainly caused by artificial planting. The results of this study can provide references for local government on mangrove management and ecological restoration.

Investigating the characteristics of dynamic changes of land use in the water source area of South-to-North Water Transfer Project (SNWTP) is of great significance to ameliorate the environmental quality to guarantee adequate, high quality water diverted into Beijing and Tianjin. The spatial and temporal characteristics of land use in the water source core area of middle route of South-to-North Water Transfer Project from 2009 to 2015 were analyzed by RS and GIS technologies based on national geoinformation survey data and basic surveying and mapping data, the results as follow: (1) the land was mainly covered with forest, which was accounting for about 80% of the total area. The area of cropland was reduced obviously, contrarily, the area of grassland increased prominently during the period from 2009 to 2015. The total coverage of wood and grass land rose lightly. (2) the land use transforming mainly occurred among cropland, woodland and grassland. Cropland was mainly changed to woodland and grassland, distributed in the altitude below 1000m and the slope above 15°. (3) the ecological engineering construction and reservoir resettlement policy had provided an effective backdrop for above changes. 53.47% of slope cropland was changed to woodland and grassland. However, there was still a considerable area of woodland, grassland transformation into slope cropland. The government should strengthen the protection of ecological environment, especially in the altitude bellow 1000m and the slope between 15° to 25°, in order to reduce the man-made environment destruction to the water supplying area.

The evaluation of small watershed management efforts is significantly meaningful to design and implement projects of protecting the water quality in water source area. In this paper, the images of GaoFen-1(GF1) and the China's first civilian high-resolution stereo mapping satellite, ZiYuan-3 (ZY-3) satellite, were used to get the land surface information in a small watershed in the water source area of middle route of South to North Water Diversion Project (SNWDP). The change of land surface was also extracted using images with different period. The effects of grain to green solutions implemented in small watershed management were then analyzed. The results indicate that high-resolution images with different periods have a great potential to monitor land cover change in small watershed and is expected to be useful for evaluating small watershed control efforts, in turn for the environment protection in water source area.

Remote sensing image is one of the most important data source for geographic information system. In 2015, China has conducted a standard point approval mission to update and improve the results of the first geographic conditions census. Majority of the geographic information was extracted from optical remote sensing image of satellite. An overall planning technique was introduced to optimize the usage of all available satellite image resources. It has shown that, only 65% of available satellite images were used to get a maximum coverage for province-scale area, and more images are needed for larger mission areas. Also, a cloud-free coverage plan of available satellite images can be designed using the overall planning technique. The overall planning technique has a great potential in designing an optimal image coverage plan over a certain area.

Remote sensing technology, geospatial analysis and statistical analysis were integrated to link land use and water quality in four river basins of Beijing. Results showed that Beijing was in a rapid urbanization process from 1989 to 2008. At buffer zone scale, area of built-up area and transportation land were positively correlated with TN and NO3--N, and negatively correlated with TP and NH4+-N; Agriculture land was strongly related to TP, while negatively related to NO3--N; other land use types were not good predictors for water quality in the study area.