Frank Zhang

Followers

Following

Co-author

Public Views

Interests

Uploads

Papers by Frank Zhang

Diversified Composite Prompting to Enhance Generalisation of Vision-Language Models

CAAI Transactions on Intelligence Technology, 2026

In recent years, prompt learning has shown promise in transferring pretrained vision-language mod... more In recent years, prompt learning has shown promise in transferring pretrained vision-language models (VLMs) to downstream tasks. However, existing methods face two challenges in improving generalisation: (1) When leveraging the collaborative effect of multimodal prompts, it is often assumed that text and visual modalities share the same prompt requirements, neglecting the distinct hierarchical processing of their encoders, leading to prompt imbalance; and (2) current methods exhibit limited adaptability when facing diverse distribution shift scenarios, including class distribution shifts and image content variations. To address these challenges, we propose a diversified composite prompt learning (DCPL) framework that integrates unified and specific prompts. Specifically, to alleviate multimodal prompt imbalance, we design a shared root multimodal prompting strategy, which employs a shared root prompt and an independent derivation mechanism to generate the derived multimodal prompt (DMP), enabling independent deep prompting while maintaining implicit synergy across modalities. Furthermore, we design a dual-branch dynamic adaptive prompting strategy that produces the derived class-specific prompt (DCP) and imagespecific prompt (ISP), driven by inter-class relations and image-patch context, respectively, to enhance adaptability across different distribution shifts. Extensive experiments on base-to-novel, cross-dataset, domain generalisation and few-shot learning demonstrate that the DCPL achieves superior performance, validating its robustness and generalisation. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

Download

Indoor Pathfinding with the A* Algorithm: A Cross-Platform Mobile Implementation Case

Journal of Advances in Engineering and Technology, 2025

This study presents the development of a fully integrated mobile module that enables indoor pathf... more This study presents the development of a fully integrated mobile module that enables indoor pathfinding functionality from the front end to the back end. The module is implemented using the A* algorithm for route optimization and a NestJS framework with PostgreSQL and PostGIS for spatial data management. Designed as part of the University of the Fraser Valley (UFV) Campus App project, this cross-platform mobile application is built with React Native to ensure seamless usability across devices. Core functionalities include intelligent room search, interactive floor plan visualization, and real-time, turn-by-turn navigation powered by WebSocket communication. The system demonstrates the feasibility of combining spatial databases, efficient routing algorithms, and real-time communication technologies to enhance campus navigation and user experience.

Download

Low-Cost CCTV Repurposing for Sustainable Parking Management: A Non-AI Computer Vision Case Study

Journal of Advances in Engineering and Technology, 2025

Urban parking inefficiency results in fuel waste and emissions, yet the current monitoring system... more Urban parking inefficiency results in fuel waste and emissions, yet the current monitoring systems rely on resource-intensive AI or sensor-based approaches. Aligning with a focus on sustainable technology, this project demonstrates how deterministic computer-vision techniques (using adaptive thresholding and polygon masking) can transform the existing CCTV camera framework into parking occupancy detectors without AI. This system deploys an open-source pipeline, combining OpenCV and PyQt5, on UFV's infrastructure, which requires zero hardware costs and consumes 95% less power than other GPU-based solutions. Testing with 18,000+ frames of simulated CCTV footage, the system achieved approximately 99% accuracy. This case study presents a replicable solution for institutions in resource-constrained environments, demonstrating that an economical IoT-CV integration can optimize urban resources while minimizing AI's carbon footprint.

Download

Individual Tree Species Classification Using Pseudo Tree Crown (PTC) on Coniferous Forests

Remote Sensing, 2025

Highlights What are the main findings? • Expanded the study of the pseudo tree crown (PTC), a pse... more Highlights What are the main findings? • Expanded the study of the pseudo tree crown (PTC), a pseudo-3D (2.5D) data representation derived from nadir UAV RGB imagery for individual tree classification and vertical structure estimation. • Demonstrated that PTC consistently improves individual conifer species classification accuracy across multiple ML/DL models compared with standard RGB images. What is the implication of the main finding? • PTC provides a cost-effective and scalable alternative to LiDAR, enabling the accurate classification of tree species and the extraction of pseudo-3D crown structure using only single nadir images. • This approach strengthens forest monitoring, biodiversity assessment, and climate resilience planning, supporting sustainable forestry and future integration with full 3D crown models.

Download

Three-Dimensional Unmanned Aerial Vehicle Path Planning in Simulated Rugged Mountainous Terrain Using Improved Enhanced Snake Optimizer (IESO

World Electric Vehicle Journal, 2025

The challenging terrain and deep ravines that characterize mountainous regions often result in sl... more The challenging terrain and deep ravines that characterize mountainous regions often result in slower path planning and suboptimal flight paths for unmanned aerial vehicles (UAVs) when traditional meta-heuristic optimization algorithms are employed. This study proposes a novel Improved Enhanced Snake Optimizer (IESO) for three-dimensional path planning and tested it in a simulated rugged mountainous terrain with obstacles and a restricted "no-fly zone". The initialization process in the enhanced snake optimizer is refined by integrating the Chebyshev chaotic map. Additionally, a non-monotonic factor is introduced to modulate the "temperature". This temperature controls the freedom of movement within the solution space. Furthermore, a boundary condition is incorporated into the dynamic opposition learning mechanism. These modifications collectively reduce the likelihood of population convergence to local optima during optimization. The feasibility of IESO is validated through time complexity and global convergence analyses. Comparative simulation experiments benchmarked IESO against five state-of-the-art biologically inspired optimization algorithms across test functions and path-planning simulated scenarios. Experimental results show that compared with five commonly used algorithms, the IESO algorithm improves the quality of flight trajectory planning by nearly 30% on average. Particularly when compared to the original SO algorithm, IESO demonstrates performance enhancement exceeding 36%, proving its superiority in UAV path planning over complex terrain.

Download

Improving Artificial-Intelligence-Based Individual Tree Species Classification Using Pseudo Tree Crown Derived from Unmanned Aerial Vehicle Imagery

Remote Sensing, 2024

Urban tree classification enables informed decision-making processes in urban planning and manage... more Urban tree classification enables informed decision-making processes in urban planning and management. This paper introduces a novel data reformation method, pseudo tree crown (PTC), which enhances the feature difference in the input layer and results in the improvement of the accuracy and efficiency of urban tree classification by utilizing artificial intelligence (AI) techniques. The study involved a comparative analysis of the performance of various machine learning (ML) classifiers. The results revealed a significant enhancement in classification accuracy, with an improvement exceeding 10% observed when high spatial resolution imagery captured by an unmanned aerial vehicle (UAV) was utilized. Furthermore, the study found an impressive average classification accuracy of 93% achieved by a classifier built on the PyTorch framework, with ResNet50 leveraged as its convolutional neural network layer. These findings underscore the potential of AI-driven approaches in advancing urban tree classification methodologies for enhanced urban planning and management practices.

Download

Evaluating the Influence of Pseudo Tree Crown (PTC) Input Alternatives for Machine Learning and Deep Learning Models on Individual Tree Classification Performance

Remote Sensing, 2026

Highlights What are the main findings? • The Pseudo Tree Crown (PTC) data transformation method c... more Highlights What are the main findings? • The Pseudo Tree Crown (PTC) data transformation method consistently improves the individual tree classification accuracy by at least 5% across various models and data variations. • Expanding the original to the green-band PTC implementation, the PTC framework exhibits robust, consistent behavior and high reliability across different data forms and transformations (including GRVI and PCA). What are the implications of the main findings? • By proving that PTC works beyond the green band, the framework can now be applied far beyond green-leaf species, unlocking its use for multiseasonal, senescent, or non-green canopy data, showcasing the advantages of PTC in terms of robustness, effectiveness, resilience, reliability, and flexibility. • PTC serves as the bridging of 2D top-down nadir imagery and true 3D structures. This injects geometric physical meaning and understanding back into classification processing, instead of strictly relying on the models' "auto-pilot".

Download

The FIFEDOM (Frequent Image Frames Enhanced Digital Orthorectified Mapping) Camera for Automatic Mapping of Tree Species and Structures

Geoscience and Remote Sensing IEEE International Symposium, 2006

The FIFEDOM (Frequent Image Frames Enhanced Digital Ortho- rectified Mapping) camera was designed... more The FIFEDOM (Frequent Image Frames Enhanced Digital Ortho- rectified Mapping) camera was designed to provide a cost-effective remote sensing method for the accurate acquisition bidirectional surface signatures, which for forest scenes is expected to yield information related to spatial distributions of individual tree species and tree structure with application in forest monitoring and management. Compared with existing regular digital cameras,

On the retrival of vegetation parameters from multi-angular hyperspectral remote sensing data

by Frank Zhang and Jianguo Wang

In this study, a new algorithm was developed to effectively use multi-angular hyperspectral remot... more In this study, a new algorithm was developed to effectively use multi-angular hyperspectral remote sensing data in the retrieval of vegetation parameters based on a coupled leaf and canopy reflectance model. Since the observations acquired at different viewing angles tend to have different noise levels, the posterior variance factors of the observations at different angles were estimated and they were then used to construct the observations' weight factors in model inversion. The developed method was validated using simulated data. The results show that the posterior variance factor can be used to characterize the uncertainty in the data from different sources and thus provides a means to weight these data in the model inversion.

On the retrival of vegetation parameters from multi-angular hyperspectral remote sensing data

by Frank Zhang and Jianguo Wang

2009 IEEE Toronto International Conference Science and Technology for Humanity (TIC-STH), 2009

Frank Zhang

Uploads

Papers by Frank Zhang

Log In