Song-Quan Ong
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
Song-Quan Ong has a total of 119 co-authors. Here are the top co-authors by number of shared publications:
- Hamdan Ahmad
Vector Control Research Unit, School of Biological Sciences, Universiti Sains Malaysia, Penang, Malaysia
9 shared publications Hamdan Ahmad
Vector Control Research Unit, Universiti Sains Malaysia, Bayan Lepas, Penang, Malaysia
5 shared publications- Abdul Hafiz Ab Majid
Household and Structural Urban Entomology Laboratory, Vector Control Research Unit, School of Biological Sciences, Universiti Sains Malaysia , Minden, Penang
4 shared publications Gomesh Nair
School of Intelligent Manufacturing Ecosystem Xi'an Jiaotong‐Liverpool University Suzhou China
4 shared publications- Toke Thomas Hoeye
Department of Ecoscience, Aarhus University, Aarhus C, Denmark
4 shared publications - Zairi Jaal
Vector Control Research Unit, School of Biological Sciences, Universiti Sains Malaysia, Penang 11800, Malaysia
4 shared publications
Latest Publications
- 4 August 2026
- Published by Oxford University Press (OUP) in Journal of Economic Entomology
- Vol. 119 (5) , 3625-3635
SolenopsisDetector : development of an automatic detection system for fire ants using computer vision and deep learning
Fire ants (Solenopsis spp. Westwood) pose a major ecological and economic threat, mainly due to the invasive potential of certain species. Current identification methods are highly dependent on taxonomic expertise, which can slow down decision-making. The development of an automated detection system could therefore support the identification process. We present SolenopsisDetector (SolenopD), an automated system for identifying Solenopsis ants using computer vision and deep learning. Following taxonomic practice, which emphasizes diagnostic body segments (head, thorax, abdomen), we evaluated 2 classification strategies: the use of whole-body images or segment-based images. To localize and crop out the target ants/segments, we first trained and compared 3 detection algorithms (YOLOv5, YOLOv8, and YOLOv11) to localize ants and their body segments. We then tested 3 classification models (ResNet, MobileNet, and InceptionV3) to categorize the ants into 4 classes. A total of 8,300 images were used to develop SolenopD. Among the detection models, whole-body detection achieved 0.931 mAP, whereas segment-based detection achieved 0.788 mAP, representing an average improvement of 18.15% using YOLOv8 and YOLOv11. However, for classification, segment-based images—especially thorax and abdomen—provided higher accuracy than head or whole-body images, with InceptionV3 performing best. To further validate our approach, we used Grad-CAM to visualize the features used in the classification. The highlighted regions corresponded to the taxonomically relevant segments and emphasized the biological interpretability of the system. In summary, SolenopD integrates YOLOv11 for precise localization of ants and body segments, followed by InceptionV3 for accurate classification of Solenopsis spp.
- 4 August 2026
- Published by Oxford University Press (OUP) in Journal of Economic Entomology
- Vol. 119 (5) , 3625-3635
SolenopsisDetector : development of an automatic detection system for fire ants using computer vision and deep learning
Fire ants (Solenopsis spp. Westwood) pose a major ecological and economic threat, mainly due to the invasive potential of certain species. Current identification methods are highly dependent on taxonomic expertise, which can slow down decision-making. The development of an automated detection system could therefore support the identification process. We present SolenopsisDetector (SolenopD), an automated system for identifying Solenopsis ants using computer vision and deep learning. Following taxonomic practice, which emphasizes diagnostic body segments (head, thorax, abdomen), we evaluated 2 classification strategies: the use of whole-body images or segment-based images. To localize and crop out the target ants/segments, we first trained and compared 3 detection algorithms (YOLOv5, YOLOv8, and YOLOv11) to localize ants and their body segments. We then tested 3 classification models (ResNet, MobileNet, and InceptionV3) to categorize the ants into 4 classes. A total of 8,300 images were used to develop SolenopD. Among the detection models, whole-body detection achieved 0.931 mAP, whereas segment-based detection achieved 0.788 mAP, representing an average improvement of 18.15% using YOLOv8 and YOLOv11. However, for classification, segment-based images—especially thorax and abdomen—provided higher accuracy than head or whole-body images, with InceptionV3 performing best. To further validate our approach, we used Grad-CAM to visualize the features used in the classification. The highlighted regions corresponded to the taxonomically relevant segments and emphasized the biological interpretability of the system. In summary, SolenopD integrates YOLOv11 for precise localization of ants and body segments, followed by InceptionV3 for accurate classification of Solenopsis spp.
- Open Access
- 15 January 2026
- Published by Oxford University Press (OUP) in Journal of Integrated Pest Management
- Vol. 17 (1)
Practical use of computer vision for cockroach monitoring in food premises: a case study in North Borneo, Malaysia
Cockroaches are major urban pests with significant medical and economic impacts and are commonly used as indicators of poor sanitation. Conventional monitoring methods, which depend on manual trap inspection and counting, are labor-intensive and limit practicality for routine surveillance. This study incorporates computer vision into cockroach monitoring by developing a deep learning–based detection workflow using YOLO object detection models. Ninety-seven food premises across 4 areas—Gaya Street, Api-Api, KK Times Square, and Damai—were surveyed using baited sticky traps. Images of captured cockroaches were processed with YOLOv5, YOLOv8, and YOLOv12 for performance comparison, with YOLOv8 showing the highest accuracy and consistency; it was subsequently used for automated detection and counting. Four cockroach species were identified: the German cockroach (Blattella germanica), American cockroach (Periplaneta americana), Australian cockroach (Validiblatta australasiae), and brown-banded cockroach (Supella longipalpa), with B. germanica accounting for 95.17% of detections. Infestation rates were highest in KK Times Square and Gaya Street, areas with dense food operations and high tourism activity. The use of computer vision enabled rapid assessment of cockroach infestations and the creation of an infestation density heatmap for risk visualization to guide authorities in monitoring infestation status.
- 15 January 2026
- Published by Oxford University Press (OUP) in Journal of Integrated Pest Management
- Vol. 17 (1)
Open AccessPractical use of computer vision for cockroach monitoring in food premises: a case study in North Borneo, Malaysia
Cockroaches are major urban pests with significant medical and economic impacts and are commonly used as indicators of poor sanitation. Conventional monitoring methods, which depend on manual trap inspection and counting, are labor-intensive and limit practicality for routine surveillance. This study incorporates computer vision into cockroach monitoring by developing a deep learning–based detection workflow using YOLO object detection models. Ninety-seven food premises across 4 areas—Gaya Street, Api-Api, KK Times Square, and Damai—were surveyed using baited sticky traps. Images of captured cockroaches were processed with YOLOv5, YOLOv8, and YOLOv12 for performance comparison, with YOLOv8 showing the highest accuracy and consistency; it was subsequently used for automated detection and counting. Four cockroach species were identified: the German cockroach (Blattella germanica), American cockroach (Periplaneta americana), Australian cockroach (Validiblatta australasiae), and brown-banded cockroach (Supella longipalpa), with B. germanica accounting for 95.17% of detections. Infestation rates were highest in KK Times Square and Gaya Street, areas with dense food operations and high tourism activity. The use of computer vision enabled rapid assessment of cockroach infestations and the creation of an infestation density heatmap for risk visualization to guide authorities in monitoring infestation status.
- 22 December 2025
- Published by Springer Nature in AI and Ethics
- Vol. 6 (1) , 80
Ethical and equitable approaches in AI for vector-borne disease management
- 22 December 2025
- Published by Springer Nature in AI and Ethics
- Vol. 6 (1) , 80
Ethical and equitable approaches in AI for vector-borne disease management
- Open Access
Bornean orangutan nest identification using computer vision and deep learning models to improve conservation strategies
Background: Regular population surveys are crucial for the evaluation of conservation measures and the management of critically endangered species such as the Bornean orangutans. Uncrewed aerial vehicles (UAV) are useful for monitoring orangutans by capturing images of the canopy, including nests, to monitor their population. However, manually detecting and counting nests from UAV imagery is time-consuming and requires trained experts. Computer vision and deep learning (DL) models for image classification offer an excellent alternative for orangutan nest identification. Methods: This study investigated DL for nest recognition from UAV imagery. A binary dataset (“with nest” and “without nest”) was created from UAV imagery from Sabah, Malaysian Borneo. The images were captured using a fixed-wing UAV with a complementary metal-oxide semiconductor camera. After image augmentation, 1,624 images were used for the dataset and further split into 70% training, 15% testing and 15% validation for model performance evaluation, i.e. , accuracy, precision, recall and F1-score. Four DL models (InceptionV3, MobileNetV2, VGG19 and Xception) were trained to learn from the labeled dataset and predict the presence of nests in new images. Results: The results show that out of four DL models, Inception V3 has the best model performance with more than 99% accuracy and precision, while VGG19 has the lowest performance. In addition, gradient-weighted class activation maps were used to interpret the results, allowing visualization of the regions used by InceptionV3 and VGG19 for classification. This study shows the potential of integrating DL into orangutan conservation, particularly in monitoring the orangutan population in the protected environment. Future research should focus on the automatic detection of nests to improve UAV-based monitoring of orangutans.
Open AccessBornean orangutan nest identification using computer vision and deep learning models to improve conservation strategies
Background: Regular population surveys are crucial for the evaluation of conservation measures and the management of critically endangered species such as the Bornean orangutans. Uncrewed aerial vehicles (UAV) are useful for monitoring orangutans by capturing images of the canopy, including nests, to monitor their population. However, manually detecting and counting nests from UAV imagery is time-consuming and requires trained experts. Computer vision and deep learning (DL) models for image classification offer an excellent alternative for orangutan nest identification. Methods: This study investigated DL for nest recognition from UAV imagery. A binary dataset (“with nest” and “without nest”) was created from UAV imagery from Sabah, Malaysian Borneo. The images were captured using a fixed-wing UAV with a complementary metal-oxide semiconductor camera. After image augmentation, 1,624 images were used for the dataset and further split into 70% training, 15% testing and 15% validation for model performance evaluation, i.e. , accuracy, precision, recall and F1-score. Four DL models (InceptionV3, MobileNetV2, VGG19 and Xception) were trained to learn from the labeled dataset and predict the presence of nests in new images. Results: The results show that out of four DL models, Inception V3 has the best model performance with more than 99% accuracy and precision, while VGG19 has the lowest performance. In addition, gradient-weighted class activation maps were used to interpret the results, allowing visualization of the regions used by InceptionV3 and VGG19 for classification. This study shows the potential of integrating DL into orangutan conservation, particularly in monitoring the orangutan population in the protected environment. Future research should focus on the automatic detection of nests to improve UAV-based monitoring of orangutans.
- 23 November 2025
- Published by Springer Nature
- p. 459-468
Comparison of Image Acquisition Devices for YOLO-Based Insect Detection: A Performance Analysis
- 23 November 2025
- Published by Springer Nature
- p. 459-468
Comparison of Image Acquisition Devices for YOLO-Based Insect Detection: A Performance Analysis
- Open Access
- 12 August 2025
- Published by Springer Nature in Magnetic Resonance Materials in Physics, Biology and Medicine
- Vol. 39 (1) , 121-135
MRI-based texture analysis for breast cancer subtype classification in a multi-ethnic population
Breast cancer, the most prevalent cancer among women globally, is classified into molecular subtypes (luminal, HER2-enriched, and triple-negative) to guide treatment and prognosis. Traditional subtyping methods, such as gene profiling and immunohistochemistry, are invasive and limited by intratumoural heterogeneity. MRI radiomics analysis offers a non-invasive alternative by extracting quantitative imaging features, yet its application in diverse, multi-ethnic populations remains underexplored. This study aimed to identify predictive radiomic features from multiple MRI sequences to classify breast cancer subtypes, compare the performance of four MRI sequences, and determine the optimal machine learning (ML) model for this task. A total of 162 retrospective breast cancer MRI cases were semi-automatically segmented, and 256 radiomic features were extracted. A multimodal ML framework integrating random forest and recursive feature elimination was developed to identify the most predictive features based on the area under the receiver operating characteristic curve (AUROC). Key predictive features included age, tumour size, margin characteristics, and intensity patterns within the tumour. Among MRI sequences, inversion recovery and T1 post-contrast performed best for subtyping. In addition, texture-based ML models effectively emulated visual assessment, demonstrating the potential of radiomics in non-invasive breast cancer subtyping. With the top ten features, the AUROC values are 0.735, 0.630, and 0.747 for luminal, HER2-enriched, and triple-negative, respectively. These findings highlight the role of MRI-based texture features and advanced ML in enhancing breast cancer diagnosis, offering a non-invasive tool for personalised treatment planning while complementing existing clinical workflows.
- 12 August 2025
- Published by Springer Nature in Magnetic Resonance Materials in Physics, Biology and Medicine
- Vol. 39 (1) , 121-135
Open AccessMRI-based texture analysis for breast cancer subtype classification in a multi-ethnic population
Breast cancer, the most prevalent cancer among women globally, is classified into molecular subtypes (luminal, HER2-enriched, and triple-negative) to guide treatment and prognosis. Traditional subtyping methods, such as gene profiling and immunohistochemistry, are invasive and limited by intratumoural heterogeneity. MRI radiomics analysis offers a non-invasive alternative by extracting quantitative imaging features, yet its application in diverse, multi-ethnic populations remains underexplored. This study aimed to identify predictive radiomic features from multiple MRI sequences to classify breast cancer subtypes, compare the performance of four MRI sequences, and determine the optimal machine learning (ML) model for this task. A total of 162 retrospective breast cancer MRI cases were semi-automatically segmented, and 256 radiomic features were extracted. A multimodal ML framework integrating random forest and recursive feature elimination was developed to identify the most predictive features based on the area under the receiver operating characteristic curve (AUROC). Key predictive features included age, tumour size, margin characteristics, and intensity patterns within the tumour. Among MRI sequences, inversion recovery and T1 post-contrast performed best for subtyping. In addition, texture-based ML models effectively emulated visual assessment, demonstrating the potential of radiomics in non-invasive breast cancer subtyping. With the top ten features, the AUROC values are 0.735, 0.630, and 0.747 for luminal, HER2-enriched, and triple-negative, respectively. These findings highlight the role of MRI-based texture features and advanced ML in enhancing breast cancer diagnosis, offering a non-invasive tool for personalised treatment planning while complementing existing clinical workflows.
- 10 August 2025
- Published by Wiley in Entomological Research
- Vol. 55 (8)
Comparison of Amino Acid Profiles for Different Deltamethrin Resistance Statuses of Aedes albopictus (Skuse)
Understanding the molecular aspects of the susceptibility status of mosquitoes to insecticides could help tackle the problem of insecticide resistance. One of the most important molecular profiles is the amino acids involved in the synthesis of metabolic resistance proteins. In this study, the amino acid profile is compared for three different deltamethrin resistance statuses of Aedes albopictus (Skuse), namely, the VCRU strain (susceptible in the laboratory), the Balik Pulau strain (susceptible in the field), and the Glugor strain (resistant in the field). The resistance status to the insecticide was determined using the WHO adulticide tube test. The mosquito samples were later freeze‐dried and derivatized by silylation and then analyzed by gas chromatography–mass spectrometry (GC–MS). Using standardized, known concentrations of amino acids, we quantified the amino acids in three different strains. A one‐way ANOVA was performed to compare the concentrations of amino acids between the strains. Of the 12 amino acids obtained from the extraction, our results showed that the susceptible strains from the laboratory and the field were not significantly different, suggesting that the profile may represent the status of susceptibility. More importantly, the resistant strain in the field had a significantly higher concentration of alanine, valine, and glycine and a significantly lower concentration of the other eight amino acids. This suggests possible metabolic adaptations related to insecticide resistance, possibly involving detoxification processes or changes in protein synthesis.
- 10 August 2025
- Published by Wiley in Entomological Research
- Vol. 55 (8)
Comparison of Amino Acid Profiles for Different Deltamethrin Resistance Statuses of Aedes albopictus (Skuse)
Understanding the molecular aspects of the susceptibility status of mosquitoes to insecticides could help tackle the problem of insecticide resistance. One of the most important molecular profiles is the amino acids involved in the synthesis of metabolic resistance proteins. In this study, the amino acid profile is compared for three different deltamethrin resistance statuses of Aedes albopictus (Skuse), namely, the VCRU strain (susceptible in the laboratory), the Balik Pulau strain (susceptible in the field), and the Glugor strain (resistant in the field). The resistance status to the insecticide was determined using the WHO adulticide tube test. The mosquito samples were later freeze‐dried and derivatized by silylation and then analyzed by gas chromatography–mass spectrometry (GC–MS). Using standardized, known concentrations of amino acids, we quantified the amino acids in three different strains. A one‐way ANOVA was performed to compare the concentrations of amino acids between the strains. Of the 12 amino acids obtained from the extraction, our results showed that the susceptible strains from the laboratory and the field were not significantly different, suggesting that the profile may represent the status of susceptibility. More importantly, the resistant strain in the field had a significantly higher concentration of alanine, valine, and glycine and a significantly lower concentration of the other eight amino acids. This suggests possible metabolic adaptations related to insecticide resistance, possibly involving detoxification processes or changes in protein synthesis.
- Open Access
- 16 May 2025
- Published by Elsevier in Data in Brief
- Vol. 60, 111673
An annotated image dataset of urban insects for the development of computer vision and deep learning models with detection tasks
- 16 May 2025
- Published by Elsevier in Data in Brief
- Vol. 60, 111673
Open AccessAn annotated image dataset of urban insects for the development of computer vision and deep learning models with detection tasks
- Open Access
- 8 May 2025
- Published by Elsevier in Current Research in Parasitology & Vector-Borne Diseases
- Vol. 7, 100268
ScannerVision: Scanner-based image acquisition of medically important arthropods for the development of computer vision and deep learning models
- 8 May 2025
- Published by Elsevier in Current Research in Parasitology & Vector-Borne Diseases
- Vol. 7, 100268
Open AccessScannerVision: Scanner-based image acquisition of medically important arthropods for the development of computer vision and deep learning models