Axiu Mao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Axiu Mao has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Dirk Udo Pfeiffer
Centre for Applied One Health Research and Policy Advice City University of Hong Kong Kowloon Hong Kong SAR P. R. China
2 shared publications - Endai Huang
Department of Computer Science City University of Hong Kong Hong Kong China
2 shared publications - Guiyuan Han
City University of Hong Kong
2 shared publications - Hsiang-Yu Yuan
Department of Biomedical Sciences, Jockey Club College of Veterinary Medicine and Life Sciences, City University of Hong Kong, Hong Kong, China
2 shared publications - Hsiangkuo Yuan
Thomas Jefferson University
2 shared publications - Kai Liu
Chongqing University
2 shared publications
Latest Publications
- Open Access
- 10 November 2023
- Published by MDPI AG in Preprints.org
Automatic Detection of Feral Pigeons in Urban Environment Using Deep Learning
The overpopulation of feral pigeons in Hong Kong has significantly disrupted the urban ecosystem, highlighting the urgent need for effective strategies to control their population. In general, control measures should be implemented and re-evaluated periodically following accurate estimations of feral pigeon population in the concerned regions, which, however, is very difficult in urban envi-ronments due to the concealment and mobility of pigeons within complex building structures. With the advances in deep learning, computer vision can be a promising tool for pigeon monitoring and population estimation but has not been well investigated so far. Therefore, we propose an improved deep learning model based on Mask-RCNN (Swin-Mask R-CNN) for feral pigeon detection using computer vision techniques. Specifically, our model consists of a Swin transformer network (STN) as the backbone, a feature pyramid network (FPN) as the neck, and three decoupled detection heads. The STN is utilized to extract deep feature information of feral pigeons through local and cross-window attention mechanisms. The FPN is employed to fuse multi-scale features and enhance the multi-scale learning ability. Heads in the three branches are responsible for classification, pre-dicting best bounding boxes, and segmentation of feral pigeons, respectively. During the prediction phase, a Slicing Aided Hyper Inference (SAHI) tool is employed to zoom in on the feature infor-mation of small feral pigeon targets, and the segmentation head is frozen to expedite inference of large images. Experiments were conducted on feral pigeon dataset to evaluate model performance. The results reveal that our model is well-suited for detecting small targets in high-resolution images and achieves excellent recognition performance for feral pigeons with a mAP (mean average pre-cision) and an AP50 (average precision at 50% intersection over union) of 0.74 and 0.93, respectively. For small target feral pigeons, AP50 in small scale (AP50s) improved by 10% as compared to the Mask R-CNN (AP50s of 0.75), demonstrating its potential for dynamic pigeon detection and population estimation in the future.
- 10 November 2023
- Published by MDPI AG in Preprints.org
Open AccessAutomatic Detection of Feral Pigeons in Urban Environment Using Deep Learning
The overpopulation of feral pigeons in Hong Kong has significantly disrupted the urban ecosystem, highlighting the urgent need for effective strategies to control their population. In general, control measures should be implemented and re-evaluated periodically following accurate estimations of feral pigeon population in the concerned regions, which, however, is very difficult in urban envi-ronments due to the concealment and mobility of pigeons within complex building structures. With the advances in deep learning, computer vision can be a promising tool for pigeon monitoring and population estimation but has not been well investigated so far. Therefore, we propose an improved deep learning model based on Mask-RCNN (Swin-Mask R-CNN) for feral pigeon detection using computer vision techniques. Specifically, our model consists of a Swin transformer network (STN) as the backbone, a feature pyramid network (FPN) as the neck, and three decoupled detection heads. The STN is utilized to extract deep feature information of feral pigeons through local and cross-window attention mechanisms. The FPN is employed to fuse multi-scale features and enhance the multi-scale learning ability. Heads in the three branches are responsible for classification, pre-dicting best bounding boxes, and segmentation of feral pigeons, respectively. During the prediction phase, a Slicing Aided Hyper Inference (SAHI) tool is employed to zoom in on the feature infor-mation of small feral pigeon targets, and the segmentation head is frozen to expedite inference of large images. Experiments were conducted on feral pigeon dataset to evaluate model performance. The results reveal that our model is well-suited for detecting small targets in high-resolution images and achieves excellent recognition performance for feral pigeons with a mAP (mean average pre-cision) and an AP50 (average precision at 50% intersection over union) of 0.74 and 0.93, respectively. For small target feral pigeons, AP50 in small scale (AP50s) improved by 10% as compared to the Mask R-CNN (AP50s of 0.75), demonstrating its potential for dynamic pigeon detection and population estimation in the future.
- Open Access
Occlusion-Resistant Instance Segmentation of Piglets in Farrowing Pens Using Center Clustering Network
Computer vision enables the development of new approaches to monitor the behavior, health, and welfare of animals. Instance segmentation is a high-precision method in computer vision for detecting individual animals of interest. This method can be used for in-depth analysis of animals, such as examining their subtle interactive behaviors, from videos and images. However, existing deep-learning-based instance segmentation methods have been mostly developed based on public datasets, which largely omit heavy occlusion problems; therefore, these methods have limitations in real-world applications involving object occlusions, such as farrowing pen systems used on pig farms in which the farrowing crates often impede the sow and piglets. In this paper, we adapt a Center Clustering Network originally designed for counting to achieve instance segmentation, dubbed as CClusnet-Inseg. Specifically, CClusnet-Inseg uses each pixel to predict object centers and trace these centers to form masks based on clustering results, which consists of a network for segmentation and center offset vector map, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, Centers-to-Mask (C2M), and Remain-Centers-to-Mask (RC2M) algorithms. In all, 4,600 images were extracted from six videos collected from three closed and three half-open farrowing crates to train and validate our method. CClusnet-Inseg achieves a mean average precision (mAP) of 84.1 and outperforms all other methods compared in this study. We conduct comprehensive ablation studies to demonstrate the advantages and effectiveness of core modules of our method. In addition, we apply CClusnet-Inseg to multi-object tracking for animal monitoring, and the predicted object center that is a conjunct output could serve as an occlusion-resistant representation of the location of an object.
Open AccessOcclusion-Resistant Instance Segmentation of Piglets in Farrowing Pens Using Center Clustering Network
Computer vision enables the development of new approaches to monitor the behavior, health, and welfare of animals. Instance segmentation is a high-precision method in computer vision for detecting individual animals of interest. This method can be used for in-depth analysis of animals, such as examining their subtle interactive behaviors, from videos and images. However, existing deep-learning-based instance segmentation methods have been mostly developed based on public datasets, which largely omit heavy occlusion problems; therefore, these methods have limitations in real-world applications involving object occlusions, such as farrowing pen systems used on pig farms in which the farrowing crates often impede the sow and piglets. In this paper, we adapt a Center Clustering Network originally designed for counting to achieve instance segmentation, dubbed as CClusnet-Inseg. Specifically, CClusnet-Inseg uses each pixel to predict object centers and trace these centers to form masks based on clustering results, which consists of a network for segmentation and center offset vector map, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, Centers-to-Mask (C2M), and Remain-Centers-to-Mask (RC2M) algorithms. In all, 4,600 images were extracted from six videos collected from three closed and three half-open farrowing crates to train and validate our method. CClusnet-Inseg achieves a mean average precision (mAP) of 84.1 and outperforms all other methods compared in this study. We conduct comprehensive ablation studies to demonstrate the advantages and effectiveness of core modules of our method. In addition, we apply CClusnet-Inseg to multi-object tracking for animal monitoring, and the predicted object center that is a conjunct output could serve as an occlusion-resistant representation of the location of an object.
- Open Access
The importance of the timing of quarantine measures before symptom onset to prevent COVID-19 outbreaks - illustrated by Hong Kong’s intervention model
Background: The rapid expansion of the current COVID-19 outbreak has caused a global pandemic but how quarantine-based measures can prevent or suppress an outbreak without other more intrusive interventions has not yet been determined. Hong Kong had a massive influx of travellers from mainland China, where the outbreak began, during the early expansion period coinciding with the Lunar New Year festival; however, the spread of the virus has been relatively limited even without imposing severe control measures, such as a full city lockdown. Understanding how quarantine measures in Hong Kong were effective in limiting community spread can provide us with valuable insights into how to suppress an outbreak. However, challenges exist in evaluating the effects of quarantine on COVID-19 transmission dynamics in Hong Kong due to the fact that the effects of border control have to be also taken into account. Methods: We have developed a two-layered susceptible-exposed-infectious-quarantined-recovered (SEIQR) meta-population model which can estimate the effects of quarantine on virus transmissibility after stratifying infections into imported and subsequent community infections, in a region closely connected to the outbreak’s source. We fitted the model to both imported and local confirmed case data with symptom onset from 18 January to 29 February 2020 in Hong Kong, together with daily transportation data and the transmission dynamics of COVID-19 from Wuhan and mainland China. After model fitting, epidemiological parameters and the timing of the start of quarantine for infected cases were estimated. Results: The model estimated that the reproduction number of COVID-19 in Hong Kong was 0.76 (95% CI, 0.66 to 0.86), achieved through quarantining infected cases −0.57 days (95% CI, −4.21 − 3.88) relative to symptom onset, with an estimated incubation time of 5.43 days (95% CI, 1.30 − 9.47). However, if delaying the quarantine start by more than 1.43 days, the reproduction number would be greater than one, making community spread more likely. The model also determined the timing of the start of quarantine necessary in order to suppress an outbreak in the presence of population immunity. Conclusion: The results suggest that the early quarantine for infected cases before symptom onset is a key factor to prevent COVID-19 outbreak.
Open AccessThe importance of the timing of quarantine measures before symptom onset to prevent COVID-19 outbreaks - illustrated by Hong Kong’s intervention model
Background: The rapid expansion of the current COVID-19 outbreak has caused a global pandemic but how quarantine-based measures can prevent or suppress an outbreak without other more intrusive interventions has not yet been determined. Hong Kong had a massive influx of travellers from mainland China, where the outbreak began, during the early expansion period coinciding with the Lunar New Year festival; however, the spread of the virus has been relatively limited even without imposing severe control measures, such as a full city lockdown. Understanding how quarantine measures in Hong Kong were effective in limiting community spread can provide us with valuable insights into how to suppress an outbreak. However, challenges exist in evaluating the effects of quarantine on COVID-19 transmission dynamics in Hong Kong due to the fact that the effects of border control have to be also taken into account. Methods: We have developed a two-layered susceptible-exposed-infectious-quarantined-recovered (SEIQR) meta-population model which can estimate the effects of quarantine on virus transmissibility after stratifying infections into imported and subsequent community infections, in a region closely connected to the outbreak’s source. We fitted the model to both imported and local confirmed case data with symptom onset from 18 January to 29 February 2020 in Hong Kong, together with daily transportation data and the transmission dynamics of COVID-19 from Wuhan and mainland China. After model fitting, epidemiological parameters and the timing of the start of quarantine for infected cases were estimated. Results: The model estimated that the reproduction number of COVID-19 in Hong Kong was 0.76 (95% CI, 0.66 to 0.86), achieved through quarantining infected cases −0.57 days (95% CI, −4.21 − 3.88) relative to symptom onset, with an estimated incubation time of 5.43 days (95% CI, 1.30 − 9.47). However, if delaying the quarantine start by more than 1.43 days, the reproduction number would be greater than one, making community spread more likely. The model also determined the timing of the start of quarantine necessary in order to suppress an outbreak in the presence of population immunity. Conclusion: The results suggest that the early quarantine for infected cases before symptom onset is a key factor to prevent COVID-19 outbreak.
- Open Access
Effectiveness of quarantine measure on transmission dynamics of COVID-19 in Hong Kong
The rapid expansion of COVID-19 has caused a global pandemic. Although quarantine measures have been used widely, the critical steps among them to suppress the outbreak without a huge social-economic loss remain unknown. Hong Kong, unlike other regions in the world, had a massive number of travellers from Mainland China during the early expansion period, and yet the spread of virus has been relatively limited. Understanding the effect of control measures to reduce the transmission in Hong Kong can improve the control of the virus spreading. We have developed a susceptible-exposed-infectious-quarantined-recovered (SEIQR) meta-population model that can stratify the infections into imported and subsequent local infections, and therefore to obtain the control effects on transmissibility in a region with many imported cases. We fitted the model to both imported and local confirmed cases with symptom onset from 18 January to 29 February 2020 in Hong Kong with daily transportation data and the transmission dynamics from Wuhan and Mainland China. The model estimated that the reproductive number was dropped from 2.32 to 0.76 (95% CI, 0.66 to 0.86) after an infected case was estimated to be quarantined half day before the symptom onset, corresponding to the incubation time of 5.43 days (95% CI, 1.30-9.47). If the quarantine happened about one day after the onset, community spread would be likely to occur, indicated by the reproductive number larger than one. The results suggest that the early quarantine for a suspected case before the symptom onset is a key factor to suppress COVID-19.
Open AccessEffectiveness of quarantine measure on transmission dynamics of COVID-19 in Hong Kong
The rapid expansion of COVID-19 has caused a global pandemic. Although quarantine measures have been used widely, the critical steps among them to suppress the outbreak without a huge social-economic loss remain unknown. Hong Kong, unlike other regions in the world, had a massive number of travellers from Mainland China during the early expansion period, and yet the spread of virus has been relatively limited. Understanding the effect of control measures to reduce the transmission in Hong Kong can improve the control of the virus spreading. We have developed a susceptible-exposed-infectious-quarantined-recovered (SEIQR) meta-population model that can stratify the infections into imported and subsequent local infections, and therefore to obtain the control effects on transmissibility in a region with many imported cases. We fitted the model to both imported and local confirmed cases with symptom onset from 18 January to 29 February 2020 in Hong Kong with daily transportation data and the transmission dynamics from Wuhan and Mainland China. The model estimated that the reproductive number was dropped from 2.32 to 0.76 (95% CI, 0.66 to 0.86) after an infected case was estimated to be quarantined half day before the symptom onset, corresponding to the incubation time of 5.43 days (95% CI, 1.30-9.47). If the quarantine happened about one day after the onset, community spread would be likely to occur, indicated by the reproductive number larger than one. The results suggest that the early quarantine for a suspected case before the symptom onset is a key factor to suppress COVID-19.