VTC 2025 Spring Conference’s Shorts

Smart Dry Bean Classification: Unleashing AI-Powered Image Analysis for Superior Precision

Dry beans are a widely consumed crop with distinct species, each possessing unique characteristics. Accurate classification is essential for quality control and efficient crop management. This study explores the multiclass classification of dry beans using various machine learning techniques, focusing on the impact of preprocessing methods—MinMax Scaler, Standard Scaler, and Robust Scaler—on model performance. Extensive experiments were conducted, with particular emphasis on the Light Gradient Boosting Machine (LGBM) classifier. Results indicate that LGBM consistently outperforms alternative models, including Multilayer Perceptron (MLP), Logistic Regression, Random Forest, K-Nearest Neighbors (KNN), Decision Tree, and Extra Tree. When utilizing the MinMax Scaler, the LGBM classifier achieved an accuracy of 96%, precision of 96%, recall of 95.80%, and an F1-score of 95.57%. These findings highlight LGBM’s effectiveness in accurately classifying dry beans while demonstrating the critical role of preprocessing techniques in optimizing model performance. Among the tested scalers, the MinMax Scaler consistently produced the highest-performing models, whereas the Standard Scaler exhibited slightly reduced performance on specific metrics. The Robust Scaler showed comparable results to the MinMax Scaler, reinforcing its suitability for handling outliers. These insights emphasize the importance of selecting an appropriate preprocessing technique based on dataset characteristics. The integration of the LGBM classifier with optimized preprocessing methods presents a powerful approach for dry bean classification, enabling precise quality assessment and informed crop management. These findings contribute to advancing machine learning applications in agriculture, offering practical guidance for researchers and industry professionals in optimizing classification models for agricultural data analysis.Smart Dry Bean Classification: Unleashing AI-Powered Image Analysis for Superior PrecisionMohamed Reda Shoeib, Jun Zhao, Nanyang Technological University

RSOF: Receiver-side Object Filtering for Scalable Collective Perception Object Fusion

Collective Perception (CP) is a key use case for the Cooperative-Intelligent Transportation System (C-ITS) to improve traffic safety. With CP, vehicles and infrastructure share information about locally detected objects via direct broadcast communication to enhance each other’s awareness and obtain perception beyond line-of-sight. However, when more and more vehicles deploy CP, the resource demand grows in terms of communication as more and more senders share their object observations. Moreover, computation requirements grow throughout a vehicle’s lifetime as more observations must be processed and fused into a holistic Environment Model. This paper proposes Receiver-Side Object Filtering (RSOF), which filters received objects by relevance and quality, reducing computing power requirements and considering constant memory. The evaluation in a large-scale traffic scenario shows that RSOF significantly reduces the number of fusion operations required to process an increasing amount of received object data. Furthermore, an upper bound for memory and fusion operations is achieved.
 
RSOF: Receiver-side Object Filtering for Scalable Collective Perception Object Fusion
 
Alexander Willecke, Fynn Schulze, Lars Wolf, TU Braunschweig, Germany

Altitude Prediction Method Using CNN-LSTM for GNSS/INS/Barometer Integrated Navigation During GNSS Outages

The global navigation satellite system (GNSS), inertial navigation system (INS), and barometer integrated navigation system is widely used in the unmanned aerial vehicles (UAVs). As UAV applications extend into complex low-altitude areas, GNSS signals may be obstructed by terrain, leading to GNSS outages. These outages are characterized by a loss of GNSS signal availability, causing a degradation of positioning accuracy under the INS/Barometer system. Therefore, GNSS measurement prediction during GNSS outages is essential. Since both the INS and barometer have limited altitude accuracy, altitude prediction becomes particularly critical. This paper establishes a GNSS measurement prediction network with separate horizontal and altitude channels. The altitude prediction channel is trained using GNSS and barometer data to improve accuracy. The network combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) to enhance the extraction of features from different sensors. Simulation results demonstrate that, during GNSS outages, positioning based on the predicted GNSS altitude using the proposed method improves altitude accuracy by 76.06% (RMSE) compared to traditional INS/Barometer-based methods.
Altitude Prediction Method Using CNN-LSTM for GNSS/INS/Barometer Integrated Navigation During GNSS Outages
 
Boyuan Liu, Zipeng Deng, Rui Xue, Beihang University

V2X Connected Smart Tyre Telemetry for Real-Time Road and Vehicle Monitoring

Vehicle-to-Everything (V2X) communication enhances road safety and traffic efficiency by supporting Cooperative Intelligent Transport System (C-ITS) services. This study introduces the Vehicle Telemetry Information Message (VTIM), designed to share real-time tyre and road surface condition data from vehicle to roadside unit (RSU). We validate the system in a field test using a connected vehicle equipped with ITS-G5 communication and smart tyre technology. Our analyses evaluate V2X latency in telemetry transmission, along with bitrate and packet error rate at the receiving end. Experimental results confirm the system’s ability to facilitate timely and reliable data exchange, enabling vehicle telemetry and road surface monitoring (e.g., water detection) for enhanced road safety services.
 
V2X Connected Smart Tyre Telemetry for Real-Time Road and Vehicle Monitoring
 
Raffaele Viterbo, Mattia Cerutti, Sanders Batista Felix, Mattia Brambilla, Politecnico di Milano; Alessandro Turati, Davide Chiola, Movyon S.p.A.; Gabriele Montorio, Pirelli Tyre S.p.A.; Monica Barbara Nicoli, Politecnico di Milano

Doppler Dilution of Precision Analysis for GNSS and Starlink LEO Satellite Positioning

In global navigation satellite system (GNSS) positioning, Doppler shift-based methods suffer from low accuracy due to the low dynamics and high altitude of medium Earth orbit (MEO) satellites. In contrast, low Earth orbit (LEO) satellites, such as Starlink, exhibit high dynamics and low altitude, leading to a significant Doppler shift effect that can enhance user terminal (UT) positioning accuracy. However, the accuracy of Doppler-based positioning is influenced mainly by the Doppler dilution of precision (DDOP) of the satellites, which quantifies the influence of satellite velocity direction and geometric diversity. Unlike conventional geometric dilution of precision (GDOP), which only quantifies the satellite’s geometric diversity, DDOP accounts for velocity vectors, making it a critical factor in Doppler-based positioning. This paper presents an intuitive analysis of DDOP and drives the Doppler geometry matrix. To evaluate the DDOP performance of Starlink satellites and GNSS, we calculate the DDOP values for evenly-spaced random places on Earth. A simulator GUI is designed to calculate the DDOP value on any place on Earth. Moreover, we establish the correlation between DDOP and Doppler-based positioning accuracy, demonstrating that lower DDOP values yield higher positioning accuracy. Our numerical analysis also confirms that Starlink LEO satellites offer significantly better DDOP than GNSS, underscoring its potential as an alternative positioning solution, particularly in the GNSS-denied environment.
 
Doppler Dilution of Precision Analysis for GNSS and Starlink LEO Satellite Positioning
 
Md. Ali Hasan, Korea University; M. Humayun Kabir, Islamic University, Bangladesh; Md. Shafiqul Islam, Bangladesh University of Business and Technology, Dhaka, Bangladesh; Takeshi Hirai, Osaka University; Wonjae Shin, Korea University

Cross-Modal Distillation by Additive Importance Measure In HITL Autonomous Driving

With the advent of Advanced Driver Assistance Systems (ADAS) and intelligent transport system applications, recognizing driver emotions has become essential for a decision support system (DSS) with humans in the loop (HITL). Multi-modal approaches using visual cues, speech, physiological signals, and driving patterns improve emotion recognition but are challenging in resource-constrained environments where only a subset of modalities is available. This work addresses these challenges by combining multi-modal benefits with single-modality inference for emotion recognition using unlabeled external road condition data. Unlike traditional methods that average teachers’ contribution, the proposed cross-modal distillation (CMD) weights teachers thanks to the Shapley additive global explanation (SAGE) aid, which improves the student model’s accuracy and provides an interpretation of it. Experimental evaluations of the PPB-Emo dataset show that XA-CMD improves emotion recognition accuracy with other baselines and provides deeper insights into decision-making.
 
Cross-Modal Distillation by Additive Importance Measure In HITL Autonomous Driving
 
Pietro Cassarà, ISTI-CNR; Saira Bano, ISTI, National Research Council (CNR); Claudio Gennaro, Information Science and Technologies (ISTI), CNR, Pisa; Alberto Gotta, ISTI-CNR

Extended vehicle energy dataset (eVED): An enhanced large-scale dataset for vehicle energy consumption analysis

This work presents an extended version of the Vehicle Energy Dataset (VED), which is a openly released large-scale dataset of vehicle trip energy consumption records. Compared with its original version, the extended VED (eVED) dataset is enhanced with accurate vehicle trip GPS coordinates. Based on the accurate trip trajectories, we associate the VED trip records with external information that is essential in analyzing vehicle energy consumption e.g., road speed limit and intersections, from open-source map services. Particularly, we calibrate all the GPS trace records in the original VED data, upon which we associated the VED data with external attributes extracted from Geographic Information System (QGIS), the Overpass API, the Open Street Map API, and Google Maps API. The extracted attributes include 12,609,170 records of road elevation, 12,203,044 of speed limit, 12,281,719 of bi-directional speed limit, 584,551 of intersections, 429,638 of bus stop, 312,196 of crossings, 195,856 of traffic signals, 29,397 of stop signs, 5,848 of turning loops, 4,053 of railway crossings, 3,554 of turning circles, and 2,938 of motorway junctions. With the accurate GPS traces and enriched features of the vehicle trip records, the obtained eVED dataset can facilitate research on vehicle energy consumption and energy-efficient approaches, especially machine/deep learning approaches that are demanding on data volume and richness. Moreover, our software work of data calibration and enrichment can be reused to generate further vehicle trip datasets for specific user cases that empower vehicle behavior and traffic dynamic analyses. We anticipate that the eVED dataset and our data enrichment software can serve the academia and industry as an apparatus in developing future vehicle technologies.Extended vehicle energy dataset (eVED): An enhanced large-scale dataset for vehicle energy consumption analysisShiliang Zhang, University of Oslo; Dyako Fatih, Fahmi Abdulqadir, Chalmers University of Technology; Tobias Schwarz, University of Gothenburg; Xuehui Ma, Xi'an University of Technology

Wideband Channel Modeling for Wireless Avionics Intra-Communications

Wireless Avionics Intra-Communications (WAIC) has been proposed to partially replace costly wiring in future generations of aircraft. Besides the allocation of a frequency range and several performance figures, there is no standardized transmission scheme so far. For the design of suitable physical layer techniques, comprehensive models are required to perform realistic link-level simulations for the aircraft environment. This paper presents a wideband channel model for in-cabin WAIC systems, derived from channel measurements performed in an Airbus A321 cabin. The model is validated by comparing experimentally measured and simulated frame error rates using the proposed model in an IEEE 802.11a-based OFDM transmission.Wideband Channel Modeling for Wireless Avionics Intra-CommunicationsJasper Brüggmann, Oscar Reyes, Christian Schappmann, Gerhard Bauch, Hamburg University of Technology

Predictive Target-to-User Association in Complex Scenarios via Hybrid-Field ISAC Signaling

This paper presents a novel and robust target-to-user (T2U) association framework to support reliable vehicle-to-infrastructure (V2I) networks that potentially operate within the hybrid field (near-field and far-field). To address the challenges posed by complex vehicle maneuvers and user association ambiguity, an interacting multiple-model filtering scheme is developed, which combines coordinated turn and constant velocity models for predictive beamforming. Building upon this foundation, a lightweight association scheme leverages user-specific integrated sensing and communication (ISAC) signaling while employing probabilistic data association to manage clutter measurements in dense traffic. Numerical results validate that the proposed framework significantly outperforms conventional methods in terms of both tracking accuracy and association reliability.Predictive Target-to-User Association in Complex Scenarios via Hybrid-Field ISAC SignalingYifeng Yuan, The Hong Kong University of Science and Technology (Guangzhou); Miaowen Wen, South China University of Technology; Xinhu Zheng, The Hong Kong University of Science and Technology (Guangzhou); Shuoyao Wang, Shenzhen University; Shijian Gao, The Hong Kong University of Science and Technology (Guangzhou)

Federated Fine-Tuning of Large Language Models for Intelligent Automotive Systems with Low...

Large Language Models (LLMs) in intelligent automotive systems offer significant benefits, such as enhancing natural language understanding, improving user interaction, and enabling more intelligent decision-making. However, this integration also faces important challenges, including data heterogeneity, limited computational resources, and the critical need to safeguard user privacy. Federated Learning (FL) offers a promising solution by enabling decentralized training across distributed data sources without compromising privacy. This paper proposes a novel FL framework for in-vehicle systems, addressing key challenges such as data heterogeneity and limited computational resources. Our method introduces a robust aggregation algorithm based on the L2 norm between LLM increments, effectively mitigating data inconsistencies and enhancing model generalization. Moreover, by integrating Low-Rank Adaptation (LoRA) within parameter-efficient fine-tuning, the framework reduces computational and communication overhead while preserving privacy. Comprehensive experiments validate that the proposed method outperforms state-of-the-art FL methods, achieving a Vicuna score of 8.17, a harmless answer rate of 68.65% (Advbenchmark), and an MTBenchmark average score of 3.74. These results highlight the potential of the proposed FL-based LLM with the LoRA framework in revolutionizing intelligent automotive systems through enhanced adaptability and privacy preservation.Federated Fine-Tuning of Large Language Models for Intelligent Automotive Systems with Low-Rank AdaptationJinhua Chen, Hosei University, Japan; Franck Junior Aboya Messou, Shilong Zhang, Hosei University; Tong Liu, Hosei University, Japan; Keping Yu, Hosei University; Dusit Niyato, Nanyang Technological University