Bingxian Wu, Yu Zhang, Zonghao Guo, Tang Liu, Chen Qian, Yuxiang Lu, Xingbo Du, Yanghao Li, Yidan Zhang, Chi Chen, Ling Yao, Chenghu Zhou, Maosong Sun. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

The growing requests for extremely fast indoor wireless connectivity have introduced significant challenges in designing next‐generation communication systems to build Internet of things (IoT)–enabled fully connected smart environments, particularly at higher frequency bands such as millimeter wave (mmWave/MMW). Accurate channel modeling is critical for optimizing these systems, especially in indoor environments where reflections, diffractions, and penetration losses considerably impact signal propagation. This study presents a detailed channel modeling approach using ray tracing techniques to characterize mmWave signal behavior in complex indoor scenarios. To accurately capture essential parameters including path loss, delay spread, and angular spread, the approach simulates signal interactions with environmental elements (e.g., walls, floors, and furniture) by leveraging three‐dimensional (3D) building models. The study provides a deeper understanding of line‐of‐sight (LOS) and non‐line‐of‐sight (NLOS) propagation. Furthermore, it comprehensively compares the propagation characteristics of various frequency bands, ranging from sub‐6 GHz (e.g., 2.4 and 6 GHz) to mmWave (e.g., 28, 60, and 100 GHz), thereby highlighting their distinct behaviors under identical indoor conditions and user trajectories. Using ray tracing, channel impulse responses and path loss metrics are extracted, and coverage map of received power is proposed for each position. Results demonstrate that mmWave bands experience higher path losses than sub‐6 GHz frequencies and are significantly affected by shadowing and blockage. This study not only validates the accuracy of the ray tracing model against empirical data but also demonstrates its utility in designing robust mmWave communication systems, optimizing network deployments, and enhancing beamforming strategies for future 5G and 6G networks.

Accurate precipitation nowcasting up to two hours in advance is crucial for weather-sensitive decision-making, such as flood disaster warnings. Although deep learning-based radar echo extrapolation methods show advantages, they still face challenges, particularly in predicting local spatial variations and mitigating performance degradation with increasing prediction time horizon. This study proposes a new multi-scale spatiotemporal prediction LSTM model (MS2T-LSTM) to enhance precipitation nowcasting capabilities by fusing multi-scale spatial features through parallel convolutional structures and improving spatiotemporal reasoning through stacked LSTMs. Notably, multi-scale spatiotemporal feature gate is introduced into LSTM unit to transmit multi-scale spatiotemporal features of radar echo image sequences, mitigating the issue of time-dependent performance degradation. The loss function for model optimization is carefully selected via series of comparative experiments. The experiments show that MS2T-LSTM outperforms benchmark models in predicting precipitation events with different intensities. Specifically, in comparison to PredRNN++, the F1 score, Critical Success Index, Heidke Skill Score, and Probability of Detection for low-intensity precipitation threshold (5mm/h) improved by 2.3%, 3.5%, 13%, and 10.6%. This study demonstrates the effectiveness of integrating multi-scale spatiotemporal features to enhance precipitation forecasting, providing valuable insights for spatiotemporal prediction modeling and optimization.

The decline of cultivated land significantly threatens the food supply. In recent years, remote sensing (RS) change detection emerged as a valuable tool for monitoring nonagriculturalization. However, existing deep learning (DL) methods for change detection suffer from the problem of inadequate utilization of feature information during image feature extraction, leading to noisy or inaccurate change maps. To address these challenges, we propose a Siamese network based on full-scale connected UNet (SNUNet3+), whose encoder extracts features from the original images at various levels. Then, it merges all fine-grained spatial information and coarse-grained semantic information into the decoder through full-scale skip connections to obtain the feature maps with complete information. The spatial and channel squeeze and excitation (scSE) attention mechanism is embedded within the decoder subnetwork to enhance the discriminative power of the features. In addition, a deep supervision module is introduced to improve the feature learning capability of the hidden layers and the quality of features. Moreover, accurate cultivated land change detection remains challenging because of the lack of fine-grained detection datasets. To enable the proposed method to achieve cultivated land change detection, we produced a cultivated land change detection dataset containing 5170 pairs of 256$\times $256 bitemporal images with a spatial resolution of 1 m. The effectiveness and performance of the proposed method are evaluated on three high-resolution RS change detection datasets. Extensive experimental results show that the proposed method outperforms other state-of-the-art methods, achieving the highest${F}1$-score of 72.90%, 90.36%, and 96.64% on the CLCD dataset, LEVIR-CD dataset, and PX-CLCD dataset, respectively.

Deep learning technologies have significantly advanced object information extraction from remote sensing data in recent years, achieving broad application across various industrial sectors. However, information loss exists between remote sensing object raster segmentation and geographic information vector mapping, making it challenging to directly apply raster extraction results to vector mapping. This study, taking the automatic extraction of greenhouses based on remote sensing imagery as an example, proposes a vectorization method for remote sensing object segmentation based on frame field. This method bridges the gap between the object pixel segmentation process and the mask vectorization process through the frame field information outputted by the network, resulting in smoother and more regular vector extraction results. To validate the effectiveness of our framework, we introduce the first high-precision greenhouse vector boundary dataset. Extensive experiments demonstrate that our method significantly mitigates the information loss issue prevalent in traditional vectorization processes, achieving a 5.05% improvement in IoU, a 6.06% increase in recall, and a 5.54% reduction in maximum angular error compared to simple vectorization schemes. It outputs more regular greenhouse vector plots, where the precision of the frame field plays a crucial role in the final vectorization quality. This research offers a unique and practical solution, converting remote sensing object segmentation into vector maps.

Tree density is a significant parameter for forest. However, it has always been challenging to use remote-sensing techniques to acquire it efficiently and effectively on a large spatial scale. As a matter of fact, tree counting can be regarded as end-to-end density regression, which uses a remote-sensing image as an input and generates a tree density map as an output. Therefore, the tree number can be easily obtained by the summation of density values in an area. To this end, a deep neural network has been constructed to aggregate multiple decoding paths to extract hierarchical features at different encoding stages, in order to merge tree features of multiple scales in remote-sensing images. At the same time, a hybrid loss function has been proposed to effectively guide the network training and enhance the model ability. A tree count dataset consisting of 2400 sample pairs has been constructed for training and validation. When compared with other popular counting networks, it has been found that the proposed network achieved the best results with a relative mean absolute error (rMAE) of 16.72%, a root mean squared error (RMSE) of 77.96, and an$R^{2}$of 0.96, evidencing that this is a promising method for estimating tree numbers on a large spatial scale.

The complexity of geographic modelling is increasing; hence, preparing data to drive geographic models is becoming a time-consuming and difficult task that may significantly hinder the application of such models. Meanwhile, a huge number of data sets have been shared and have become publicly accessible through the Internet. This study presents a data similarity-based approach to automatically recommend available data sets to fulfil the data requirements of geographic models. Unified description factors are adopted to provide a consistent description of public data sets and input data requirements of geographic models. Five elementary data similarities between them, specifically content, spatial coverage, temporal coverage, spatial precision, and temporal granularity similarities, are calculated. An overall similarity is estimated from aggregating the elementary data similarities. Thereafter, the candidate data for running the models are recommended in the order of overall data similarity. As a case study, the approach has been applied to recommend data from the China National Data Sharing Platform of Earth System Science to drive the population spatialization model (PSM). The approach has successfully recommended the most related data sets to run PSM. The result also suggests that the data recommendation approach can facilitate the intelligent identification of geographic data and the building of links between the open data sets.

Unlike wired networks, random packet loss due to bit errors may cause significant performance degradation of the transmission control protocol (TCP). We propose and study a novel end-to-end congestion control mechanism called TCP/spl I.bar/LD (lose detection) that is simple and effective for dealing with random packet loss . The simulation results show that our scheme can achieve significant throughput improvements without adversely affecting other concurrent TCP connections, including other concurrent Reno connections.