Ziyi Sun
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ziyi Sun has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Bing Xue
Centre for Data Science and Artificial Intelligence & School of Engineering and Computer Science, Victoria University of Wellington, New Zealand
3 shared publications Jan Schindler
Manaaki Whenua-Landcare Research
3 shared publications- Mengjie Zhang
Centre for Data Science and Artificial Intelligence & School of Engineering and Computer Science, Victoria University of Wellington, New Zealand
3 shared publications Qun Jiang
Affiliation School of Computer Science and Technology, Shandong University of Finance and Economics, Jinan, China
2 shared publicationsBing Xue
Center for Data Science and Artificial Intelligence & School of Engineering and Computer Science Victoria University of Wellington Wellington New Zealand
1 shared publication- Fangxun Bao
School of Mathematics, Shandong University, Jinan, China
1 shared publication
Latest Publications
- 1 May 2026
- Published by Springer Nature
- p. 298-314
YOLOv8-UT: A Unified Training Approach for Cross-Environment Tree Crown Instance Segmentation in Aerial Imagery
- 1 May 2026
- Published by Springer Nature
- p. 298-314
YOLOv8-UT: A Unified Training Approach for Cross-Environment Tree Crown Instance Segmentation in Aerial Imagery
- Top Cited Papers
- 1 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 10 (1) , 51-65
YOLO-ITC: A New YOLO Method for Instance Segmentation of Individual Tree Crowns
The instance segmentation task of individual tree crowns is an important real-world application that facilitates forest management, carbon storage estimation, and biodiversity modelling. Recently, Convolutional Neural Networks (CNNs) have achieved great success in computer vision. Several efforts have applied CNNs to perform instance segmentation of tree canopies. However, the existing CNN works rarely consider instance segmentation and species classification of densely distributed small and medium tree crowns. YOLO, as a fast and effective method, shows promising results in the field of instance segmentation and canopy detection and segmentation. We utilize a tree crown dataset from Wellington, Aotearoa, New Zealand, based on aerial imagery which contains small and medium-sized tree objects distributed across a diverse rural landscape ranging from isolated trees to dense forest stands. We propose a new YOLO method for individual tree crowns (ITC) based on YOLOv7, named YOLO-ITC. Specifically, a new detection design targeting small and medium trees is first proposed. Based on this, we propose a Dense Block module for the main block of the backbone in a densely connected way, where the previous features are reused to aggregate more useful information. Finally, efficient attention is embedded in the Dense Block to capture the globally long-range dependencies. Extensive experimental results show that the proposed method, YOLO-ITC, achieves both higher detection and mask accuracy over other commonly used baseline models.
- 1 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Emerging Topics in Computational Intelligence
- Vol. 10 (1) , 51-65
Top Cited PapersYOLO-ITC: A New YOLO Method for Instance Segmentation of Individual Tree Crowns
The instance segmentation task of individual tree crowns is an important real-world application that facilitates forest management, carbon storage estimation, and biodiversity modelling. Recently, Convolutional Neural Networks (CNNs) have achieved great success in computer vision. Several efforts have applied CNNs to perform instance segmentation of tree canopies. However, the existing CNN works rarely consider instance segmentation and species classification of densely distributed small and medium tree crowns. YOLO, as a fast and effective method, shows promising results in the field of instance segmentation and canopy detection and segmentation. We utilize a tree crown dataset from Wellington, Aotearoa, New Zealand, based on aerial imagery which contains small and medium-sized tree objects distributed across a diverse rural landscape ranging from isolated trees to dense forest stands. We propose a new YOLO method for individual tree crowns (ITC) based on YOLOv7, named YOLO-ITC. Specifically, a new detection design targeting small and medium trees is first proposed. Based on this, we propose a Dense Block module for the main block of the backbone in a densely connected way, where the previous features are reused to aggregate more useful information. Finally, efficient attention is embedded in the Dense Block to capture the globally long-range dependencies. Extensive experimental results show that the proposed method, YOLO-ITC, achieves both higher detection and mask accuracy over other commonly used baseline models.
- Open Access
- 5 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 14752-14765
Feature Fusion to Improve YOLOv8 for Segmenting and Classifying Aerial Images of Tree Crowns
Instance segmentation techniques based on Convolutional Neural Networks (CNNs) is a vital tool for accurately identifying and segmenting individual tree crowns, which plays an essential role in environmental monitoring and forest management. In varied rural landscapes, canopy imagery often includes a mix of tiny, small, and medium tree objects scattered across diverse terrains, from standalone trees to densely clustered forest stands. This variability poses significant challenges to traditional instance segmentation methods. To achieve this, we introduce a new method named YOLOv8-FF, which incorporates a feature fusion (FF) technique based on the YOLOv8 architecture. We first design a network architecture based on YOLOv8 that is optimized for the characteristics of our dataset, enabling effective segmentation of densely distributed tiny and small tree crowns. Moreover, YOLOv8-FF incorporates a feature fusion mechanism that includes both cross-scale and same-scale fusion methods, enhancing the model's ability to integrate information across different layers and scales, thereby improving segmentation performance. We incorporate Sparse Large Kernel Network (SLaK), whose large convolution kernel can effectively extract key features, helping the model capture richer and deeper global information in the image. Experimental results on the tree crown dataset demonstrate that YOLOv8-FF outperforms several recent peer competitors, making it a promising tool for accurate and efficient tree crown instance segmentation.
- 5 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 14752-14765
Open AccessFeature Fusion to Improve YOLOv8 for Segmenting and Classifying Aerial Images of Tree Crowns
Instance segmentation techniques based on Convolutional Neural Networks (CNNs) is a vital tool for accurately identifying and segmenting individual tree crowns, which plays an essential role in environmental monitoring and forest management. In varied rural landscapes, canopy imagery often includes a mix of tiny, small, and medium tree objects scattered across diverse terrains, from standalone trees to densely clustered forest stands. This variability poses significant challenges to traditional instance segmentation methods. To achieve this, we introduce a new method named YOLOv8-FF, which incorporates a feature fusion (FF) technique based on the YOLOv8 architecture. We first design a network architecture based on YOLOv8 that is optimized for the characteristics of our dataset, enabling effective segmentation of densely distributed tiny and small tree crowns. Moreover, YOLOv8-FF incorporates a feature fusion mechanism that includes both cross-scale and same-scale fusion methods, enhancing the model's ability to integrate information across different layers and scales, thereby improving segmentation performance. We incorporate Sparse Large Kernel Network (SLaK), whose large convolution kernel can effectively extract key features, helping the model capture richer and deeper global information in the image. Experimental results on the tree crown dataset demonstrate that YOLOv8-FF outperforms several recent peer competitors, making it a promising tool for accurate and efficient tree crown instance segmentation.
- 17 November 2024
- Published by Wiley in Journal of the Royal Society of New Zealand
- Vol. 55 (6) , 1854-1879
YOLOv8E: an efficient YOLOv8 method for instance segmentation of individual tree crowns in Wellington City, New Zealand
Instance segmentation is crucial for analysing individual tree crowns in aerial imagery, which plays an important role in forest management, risk modelling, biodiversity modelling, forest health and human wellbeing studies. Traditional instance segmentation methods struggle in diverse rural landscapes where canopy images feature primarily small and medium tree objects, varying from isolated trees to dense forest stands. This paper introduces YOLOv8E, a new and efficient YOLOv8 method, optimised for precise instance segmentation and species classification of tree crowns. This method includes new schemes for selecting candidate positive samples for each instance and a refined network design tailored for small and medium‐sized tree crowns. Adjustments in hyperparameters, particularly within the Task‐Aligned Assigner, are also discussed to better suit canopy segmentation tasks. Comprehensive experiments conducted on the datasets for Wellington City, Aotearoa New Zealand, demonstrate that YOLOv8E outperforms a number of recent methods, achieving 36.1 and 32.2 in terms of the Box AP and Mask AP metrics respectively.
- 17 November 2024
- Published by Wiley in Journal of the Royal Society of New Zealand
- Vol. 55 (6) , 1854-1879
YOLOv8E: an efficient YOLOv8 method for instance segmentation of individual tree crowns in Wellington City, New Zealand
Instance segmentation is crucial for analysing individual tree crowns in aerial imagery, which plays an important role in forest management, risk modelling, biodiversity modelling, forest health and human wellbeing studies. Traditional instance segmentation methods struggle in diverse rural landscapes where canopy images feature primarily small and medium tree objects, varying from isolated trees to dense forest stands. This paper introduces YOLOv8E, a new and efficient YOLOv8 method, optimised for precise instance segmentation and species classification of tree crowns. This method includes new schemes for selecting candidate positive samples for each instance and a refined network design tailored for small and medium‐sized tree crowns. Adjustments in hyperparameters, particularly within the Task‐Aligned Assigner, are also discussed to better suit canopy segmentation tasks. Comprehensive experiments conducted on the datasets for Wellington City, Aotearoa New Zealand, demonstrate that YOLOv8E outperforms a number of recent methods, achieving 36.1 and 32.2 in terms of the Box AP and Mask AP metrics respectively.
- Open Access
- 30 July 2021
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 16 (7) , e0254664
Single image mixed dehazing method based on numerical iterative model and DehazeNet
As one of the most common adverse weather phenomena, haze has caused detrimental effects on many computer vision systems. To eliminate the effect of haze, in the field of image processing, image dehazing has been studied intensively, and many advanced dehazing algorithms have been proposed. Physical model-based and deep learning-based methods are two competitive methods for single image dehazing, but it is still a challenging problem to achieve fidelity and effectively dehazing simultaneously in real hazy scenes. In this work, a mixed iterative model is proposed, which combines a physical model-based method with a learning-based method to restore high-quality clear images, and it has good performance in maintaining natural attributes and completely removing haze. Unlike previous studies, we first divide the image into different regions according to the density of haze to accurately calculate the atmospheric light for restoring haze-free images. Then, dark channel prior and DehazeNet are used to jointly estimate the transmission to promote the final clear haze-free image that is more similar to the real scene. Finally, a numerical iterative strategy is employed to further optimize the atmospheric light and transmission. Extensive experiments demonstrate that our method outperforms existing state-of-the-art methods on synthetic datasets and real-world datasets. Moreover, to indicate the universality of the proposed method, we further apply it to the remote sensing datasets, which can also produce visually satisfactory results.
- 30 July 2021
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 16 (7) , e0254664
Open AccessSingle image mixed dehazing method based on numerical iterative model and DehazeNet
As one of the most common adverse weather phenomena, haze has caused detrimental effects on many computer vision systems. To eliminate the effect of haze, in the field of image processing, image dehazing has been studied intensively, and many advanced dehazing algorithms have been proposed. Physical model-based and deep learning-based methods are two competitive methods for single image dehazing, but it is still a challenging problem to achieve fidelity and effectively dehazing simultaneously in real hazy scenes. In this work, a mixed iterative model is proposed, which combines a physical model-based method with a learning-based method to restore high-quality clear images, and it has good performance in maintaining natural attributes and completely removing haze. Unlike previous studies, we first divide the image into different regions according to the density of haze to accurately calculate the atmospheric light for restoring haze-free images. Then, dark channel prior and DehazeNet are used to jointly estimate the transmission to promote the final clear haze-free image that is more similar to the real scene. Finally, a numerical iterative strategy is employed to further optimize the atmospheric light and transmission. Extensive experiments demonstrate that our method outperforms existing state-of-the-art methods on synthetic datasets and real-world datasets. Moreover, to indicate the universality of the proposed method, we further apply it to the remote sensing datasets, which can also produce visually satisfactory results.
- Open Access
- 20 April 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 63685-63699
A Mixed Transmission Estimation Iterative Method for Single Image Dehazing
Considering the frequent occurrence of hazy weather, single image dehazing has become an important research task. Physical model-based and deep learning-based methods are two competitive methods in single image dehazing, but achieving fidelity and effectively removing haze at the same time in real hazy scenes is still a challenging problem. In this work, we propose a mixed iterative model to restore high-quality clear images by integrating a physical model-based approach and a learning-based approach, which can combine the advantages of the physical model-based method in maintaining natural attributes and the advantages of the learning-based method in completely removing haze. For a hazy image, first, according to the haze density, it is divided into separate regions to calculate the local atmospheric light. Then, we utilize the dark channel prior and DehazeNet to jointly estimate the transmission, which can estimate an accurate recovered image that is more in line with the real scene. Finally, a numerical iterative strategy is employed to further optimize the atmospheric light and transmission. The experiments on both synthetic and natural hazy image datasets indicate that the proposed method can restore natural and clean images and has superior performance than other state-of-the-art algorithms in peak signal-to-noise ratio, structural similarity, information entropy, color natural index, and color histogram. And in the four indicators of the quantitative results of the synthetic hazy test set, we achieve the highest score in three of them and the second highest result in another. In the comparison of the two quantitative indicators of natural hazy images, our average objective results are both the highest.
- 20 April 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 63685-63699
Open AccessA Mixed Transmission Estimation Iterative Method for Single Image Dehazing
Considering the frequent occurrence of hazy weather, single image dehazing has become an important research task. Physical model-based and deep learning-based methods are two competitive methods in single image dehazing, but achieving fidelity and effectively removing haze at the same time in real hazy scenes is still a challenging problem. In this work, we propose a mixed iterative model to restore high-quality clear images by integrating a physical model-based approach and a learning-based approach, which can combine the advantages of the physical model-based method in maintaining natural attributes and the advantages of the learning-based method in completely removing haze. For a hazy image, first, according to the haze density, it is divided into separate regions to calculate the local atmospheric light. Then, we utilize the dark channel prior and DehazeNet to jointly estimate the transmission, which can estimate an accurate recovered image that is more in line with the real scene. Finally, a numerical iterative strategy is employed to further optimize the atmospheric light and transmission. The experiments on both synthetic and natural hazy image datasets indicate that the proposed method can restore natural and clean images and has superior performance than other state-of-the-art algorithms in peak signal-to-noise ratio, structural similarity, information entropy, color natural index, and color histogram. And in the four indicators of the quantitative results of the synthetic hazy test set, we achieve the highest score in three of them and the second highest result in another. In the comparison of the two quantitative indicators of natural hazy images, our average objective results are both the highest.