Papers - THI THI ZIN
-
Behavior Estimation of Calf Groups Using RGB Cameras and Deep Learning Reviewed International journal
D. Nishimoto, Thi Thi Zin., M. Aikawa
Institute of Electrical and Electronics Engineers Inc., Conference Proceedings (ICCE-Taiwan 2025) 115 - 116 2025.10
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Icce Taiwan 2025 12th IEEE International Conference on Consumer Electronics Taiwan Generative AI in Innovative Consumer Technology Proceedings
This paper presents a real-time behavior estimation system for calf groups using RGB cameras and deep learning. The system employs YOLO for detection, Weighted Intersection over Union (IoU) based tracking for consistent IDs, and Segment Anything Model2 (SAM2) with EfficientNetv2-L for individual identification. It classifies postures (sitting, standing) and intake behaviors (drinking milk/water, eating) for comprehensive health monitoring. Experiments on 16 calves achieved 91.33% Multi object tracking accuracy (MOTA), approximately 80% accuracy for posture classification, and 50-70% accuracy for intake behaviors, with real-time processing. The system effectively reduces labor burdens and supports scalable livestock management.
DOI： 10.1109/ICCE-Taiwan66881.2025.11207893
Other Link： https://ieeexplore.ieee.org/document/11207893
-
A Study on Machine Learning Approaches for Predicting Fetal pH Level Using Fetal Heart Rate Variability Reviewed International journal
Cho Nilar Phyo, Tunn Cho Lwin, Pyae Phyo Kyaw, E. Kino, T. Ikenoue, Pyke Tin, Thi Thi Zin
ICIC Express Letters, Part B: Applications 16 ( 8 ) 879 - 886 2025.8
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：Icic Express Letters Part B Applications
Fetal well-being monitoring system is essential for ensuring healthy labor outcomes. One of the non-invasive methods for assessing fetal health during labor and delivery is by analyzing fetal heart rate variability (FHRV), which can be used to predict fetal pH levels. This study compares different machine learning approaches for predicting fetal pH levels based on FHRV data collected during labor and delivery. The dataset used in this study includes FHRV signals together with corresponding umbilical cord blood gas measurements such as pH, which are used to train and evaluate the models. This study applies several machine learning algorithms and evaluates their performance using key metrics such as sensitivity, specificity, precision, F1-score, and accuracy. These metrics help to determine which model is the most accurate predicting fetal pH levels based on FHRV characteristics. The results reveal that the support vector machine (SVM) model outperforms with the accuracy of 81.67%, better than the other algorithms in predicting fetal pH levels. The findings of this study aim to contribute to the development of more reliable and accurate prediction models for assessing fetal well-being during labor, enhanced clinical decision-making, allowing for timely interventions and improved healthy labor outcomes for both the mother and the baby.
-
Machine Learning-Based Prediction of Cattle Body Condition Score using 3D Point Cloud Surface Features Reviewed International journal
Pyae Phyo Kyaw, Thi Thi Zin, Pyke Tin, M. Aikawa, I. Kobayashi
Proceedings of SPIE the International Society for Optical Engineering 13701 2025.7
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Proceedings of SPIE the International Society for Optical Engineering
Body Condition Score (BCS) of dairy cattle is a crucial indicator of their health, productivity, and reproductive performance throughout the production cycle. Recent advancements in computer vision techniques have led to the development of automated BCS prediction systems. This paper proposes a BCS prediction system that leverages 3D point cloud surface features to enhance accuracy and reliability. Depth images are captured from a top-view perspective and processed using a hybrid depth image detection model to extract the cattle’s back surface region. The extracted depth data is converted into point cloud data, from which various surface features are analyzed, including normal vectors, curvature, point density, and surface shape characteristics (planarity, linearity, and sphericity). Additionally, Fast Point Feature Histograms (FPFH), triangle mesh area, and convex hull area are extracted and evaluated using three optimized machine learning models: Random Forest (RF), K-Nearest Neighbors (KNN), and Gradient Boosting (GB). Model performance is assessed using different tolerance levels and error metrics, including Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Among the models, Random Forest demonstrates the highest performance, achieving accuracy rates of 51.36%, 86.21%, and 97.83% at 0, 0.25, and 0.5 tolerance levels, respectively, with an MAE of 0.161 and MAPE of 5.08%. This approach enhances the precision of BCS estimation, offering a more reliable and automated solution for dairy cattle monitoring and health management.
DOI： 10.1117/12.3070481
-
Minimizing Resource Usage for Real-Time Network Camera Tracking of Black Cows Reviewed International journal
Aung Si Thu Moe, Thi Thi Zin, Pyke Tin, M. Aikawa, I. Kobayashi
Proceedings of SPIE the International Society for Optical Engineering 13701 2025.7
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Proceedings of SPIE the International Society for Optical Engineering
Livestock plays a crucial role in the farming industry to meet consumer demand. A livestock monitoring system helps track animal health while reducing labor requirements. Most livestock farms are small, family-owned operations. This study proposes a real-time black cow detection and tracking system using network cameras in memory and disk constrained environments. We employ the Detectron2 Mask R-CNN ResNeXt-101 model for black cow region detection and the ByteTrack algorithm for tracking. ByteTrack tracks multiple objects by associating each detection box. Unlike other deep learning tracking algorithms that use multiple features such as texture, color, shape, and size. ByteTrack effectively reduces tracking ID errors and ID switches. Detecting and tracking black cows in real-time is challenging due to their uniform color and similar sizes. To optimize performance on low-specification machines, we apply ONNX (Open Neural Network Exchange) to the Detectron2 detection model for optimization and quantization. The system processes input images from network cameras, enhances color during preprocessing, and detects and tracks black cows efficiently. Our system achieves 95.97% mAP@0.75 detection accuracy and 97.16 % in daytime video and 94.83 % in nighttime accuracy of tracking are effectively tracks individual black cows, minimizing duplicate IDs and improving tracking after missed detections or occlusions. The system is designed to operate on machines with minimal hardware requirements.
DOI： 10.1117/12.3070347
-
Automatic cattle identification system based on color point cloud using hybrid PointNet++ Siamese network Reviewed International journal
Pyae Phyo Kyaw, Pyke Tin, M. Aikawa, I. Kobayashi, Thi Thi Zin
Scientific reports 15 ( 1 ) 21938 (article number) 2025.7
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：Nature Research, Scientific Reports
Cattle health monitoring and management systems are essential for farmers and veterinarians, as traditional manual health checks can be time-consuming and labor-intensive. A critical aspect of such systems is accurate cattle identification, which enables effective health monitoring. Existing 2D vision-based identification methods have demonstrated promising results; however, their performance is often compromised by environmental factors, variations in cattle texture, and noise. Moreover, these approaches require model retraining to recognize newly introduced cattle, limiting their adaptability in dynamic farm environments. To overcome these challenges, this study presents a novel cattle identification system based on color point clouds captured using RGB-D cameras. The proposed approach employs a hybrid detection method that first applies a 2D depth image detection model before converting the detected region into a color point cloud, allowing for robust feature extraction. A customized lightweight tracking approach is implemented, leveraging Intersection over Union (IoU)-based bounding box matching and mask size analysis to consistently track individual cattle across frames. The identification framework is built upon a hybrid PointNet ++ Siamese Network trained with a triplet loss function, ensuring the extraction of discriminative features for accurate cattle identification. By comparing extracted features against a pre-stored database, the system successfully predicts cattle IDs without requiring model retraining. The proposed method was evaluated on a dataset consisting predominantly of Holstein cow along with a few Jersey cows, achieving an average identification accuracy of 99.55% over a 13-day testing period. Notably, the system can successfully detect and identify unknown cattle without requiring model retraining. This cattle identification research aims to integrate the comprehensive cattle health monitoring system, encompassing lameness detection, body condition score evaluation, and weight estimation, all based on point cloud data and deep learning techniques.
-
A Study on the Analysis and Classification of Gait States Using Keypoint Information Reviewed International journal
Ryusei Tanno, Thi Thi Zin, Cho Nilar Phyo
ICIC Express Letters 19 ( 6 ) 677 - 684 2025.6
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters
In Japan, 29.0% of the population is elderly, and as this demographic grows, the need for care increases, straining healthcare facilities. Mobility issues, especially falls, significantly contribute to this demand. This study aims to provide a non-contact, accessible method for quantifying gait states in elderly individuals using image processing technology. The experiment, conducted at a commercial facility in Higashi-Osaka City, involved capturing walking sequences from two viewpoints with RGB cameras. Using OpenPose to extract skeletal keypoints, walking balance and kyphosis were evaluated. The angle between the neck and hips served as an indicator for balance, classified as “Normal”, “Warning”, or “Danger”, while kyphosis was classified as “Normal”, “Mild”, or “Severe”. Results showed a correlation between age and balance decline, with older individuals having more “Danger” classifications. Kyphosis was also accurately identified through visual posture comparison.
DOI： 10.24507/icicel.19.06.677
Other Link： https://www.scopus.com/pages/publications/105007210208?origin=resultslist
-
Advanced Predictive Analytics for Fetal Heart Rate Variability Using Digital Twin Integration. Reviewed International journal
Tunn Cho Lwin, Thi Thi Zin, Pyke Tin, E. Kino, T. Ikenoue
Sensors (Basel, Switzerland) 25 ( 5 ) 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：Sensors
Fetal heart rate variability (FHRV) is a critical indicator of fetal well-being and autonomic nervous system development during labor. Traditional monitoring methods often provide limited insights, potentially leading to delayed interventions and suboptimal outcomes. This study proposes an advanced predictive analytics approach by integrating approximate entropy analysis with a hidden Markov model (HMM) within a digital twin framework to enhance real-time fetal monitoring. We utilized a dataset of 469 fetal electrocardiogram (ECG) recordings, each exceeding one hour in duration, to ensure sufficient temporal information for reliable modeling. The FHRV data were preprocessed and partitioned into parasympathetic and sympathetic components based on downward and non-downward beat detection. Approximate entropy was calculated to quantify the complexity of FHRV patterns, revealing significant correlations with umbilical cord blood gas parameters, particularly pH levels. The HMM was developed with four hidden states representing discrete pH levels and eight observed states derived from FHRV data. By employing the Baum–Welch and Viterbi algorithms for training and decoding, respectively, the model effectively captured temporal dependencies and provided early predictions of the fetal acid–base status. Experimental results demonstrated that the model achieved 85% training and 79% testing accuracy on the balanced dataset distribution, improving from 78% and 71% on the imbalanced dataset. The integration of this predictive model into a digital twin framework offers significant benefits for timely clinical interventions, potentially improving prenatal outcomes.
DOI： 10.3390/s25051469
-
Optimizing black cattle tracking in complex open ranch environments using YOLOv8 embedded multi-camera system. Reviewed International journal
Su Myat Noe, Thi Thi Zin, I. Kobayashi, Pyke Tin
Scientific reports 15 ( 1 ) 6820 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：Scientific Reports
Monitoring the daily activity levels of black cattle is a crucial aspect of their well-being. The rapid advancements in artificial intelligence have transformed computer vision applications, including object detection, segmentation, and tracking. This has led to more effective and precise monitoring techniques for livestock. In modern cattle farms, video monitoring is essential for analyzing behavior, evaluating health, and predicting estrus events in precision farming. This paper introduces the novel Customized Multi-Camera Multi-Cattle Tracking (MCMCT) system. This unique approach uses four cameras to overcome the challenges of detecting and tracking black cattle in complex open ranch environments. The MCMCT system enhances a tracking-by-detection model with the YOLO v8 segmentation model as the detection backbone network to develop a precision black cattle monitoring system. Single-camera setups in real-world datasets of our open ranches, covering 23.3 m x 20 m with 55 cattle, have limitations in capturing all necessary details. Therefore, a multi-camera solution provides better coverage and more accurate behavior detection of cattle. The effectiveness of the MCMCT system is demonstrated through experimental results, with the YOLOv8-MCMCT system achieving an average Multi-Object Tracking Accuracy (MOTA) of 95.61% across 10 cases of 4 cameras at a processing speed of 30 frames per second. This high accuracy is a testament to the performance of the proposed MCMCT system. Additionally, integrating the Segment Anything Model (SAM) with YOLOv8 enhances the system’s capability by automating cattle mask region extraction, reducing the need for manual labeling. Comparative analysis with state-of-the-art deep learning-based tracking methods, including Bot-sort, Byte-track, and OC-sort, further highlights the MCMCT’s performance in multi-cattle tracking within complex natural scenes. The advanced algorithms and capabilities of the MCMCT system make it a valuable tool for non-contact automatic livestock monitoring in precision cattle farming. Its adaptability ensures effective performance across varied ranch environments without extensive retraining. This research significantly contributes to livestock monitoring, offering a robust solution for tracking black cattle and enhancing overall agricultural efficiency and management.
-
Utilizing Behavioral Features for Predicting Calving Time Reviewed International journal
Wai Hnin Eaindrar Mg, Pyke Tin, M. Aikawa, I. Kobayashi, Y. Horii, K. Honkawa K., Thi Thi Zin
Lecture Notes in Electrical Engineering 1321 LNEE 148 - 159 2025.2
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
Accurately predicting calving time in cattle is crucial for enhancing livestock management and ensuring animal welfare. Our research presents a novel approach combining advanced computer vision and deep learning techniques to predict calving time based on cattle behavior. We employ a custom YOLOv8 model for cattle detection, achieving robust and efficient localization of individual cattle in various farm environments. Our Customized Tracking Algorithm (CTA) is utilized to maintain continuous identity tracking for each cow, allowing for precise monitoring of behavioral patterns. Feature extraction is performed using ResNet50, capturing detailed spatial characteristics of the detected cattle. These features are then organized into sequences to prepare them for temporal analysis. Finally, Custom Long Short-Term Memory (CLSTM) network is used for classification, leveraging the sequential nature of the data to predict the onset of calving with high accuracy. Our classification approach achieved an average accuracy of 94.88%. Our findings indicate promising performance from our CLSTM algorithm, accurately forecasting the remaining 3 h before calving. Through a comprehensive exploration of data collection, pre-processing, and feature engineering, our research paper establishes the foundation for training an accurate behavior model to predict calving time. Predicting calving using traditional, manual methods like observing breeding records and visual cues is complex and prone to errors, with even experts sometimes failing to make accurate predictions. Additionally, manual prediction becomes impractical and costly as farm size increases. Our automated system demonstrated a significant improvement in prediction accuracy, reducing false positives and providing timely alerts. Our proposed method demonstrates significant potential for improving the precision and reliability of calving time predictions, offering valuable insights for farm management and veterinary care.
-
A Mathematical Framework for Reinforcement Learning in Healthcare: Modeling and Analysis in Artificial Intelligence Reviewed International journal
Cho Nilar Phyo,Thi Thi Zin, H. Hama, Pyke Tin
Lecture Notes in Electrical Engineering 1321 LNEE 54 - 62 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
A Reinforcement Learning, a pivotal component of artificial intelligence, is employed by computers to learn intelligently. This study delves into the application of a significant mathematical concept, the Markov Decision Process (MDP), within the realm of Reinforcement Learning. The primary focus lies in the development of a proficient computer program designed to tackle healthcare issues. The proposed approach consists of two key elements: a unique decision-making framework and intelligent learning mechanisms. This process of combining two elements is leveraged to analyze patient information and ascertain optimal choices. Conceptualizing a patient’s healthcare journey as distinct states—such as physician visits (O), hospitalization (H), intensive care (I), or mortality (D)—the research formulates a Markov Chain Model. This model quantifies the transition probabilities between these states. Additionally, an auxiliary model is constructed to gauge the efficacy of decisions, encompassing factors like risk assessment and potential medication outcomes. The effectiveness of the proposed model, termed the Markov Model with Reinforcement Learning, is evaluated using real-world patient data from electronic health records. Encouragingly, the model demonstrates proficiency in predicting forthcoming healthcare events. This underscores its utility in prognosticating future developments within the healthcare domain.
-
A Study on Health Management by Behavior Analysis of Calves Reviewed International journal
T. Nishiyama, S. Kazuhisa, M. Aikawa, I. Kobayashi, Thi Thi Zin
Lecture Notes in Electrical Engineering 1322 LNEE 144 - 151 2025.2
Authorship：Last author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
It is important to always monitor the health of cattle, especially calves, and the frequency of observation increases with environmental changes in addition to once a day. In addition, calves tend to be more susceptible to infectious diseases because of their immature immune systems. Therefore, rearing management is extremely important. And the number of dairy cattle-keeping households and the total number of cattle are decreasing, while the number of cattle per household is increasing, indicating that management is becoming larger in scale. In this study, we proposed the development of a health management system by analyzing calf behavior using a 3D camera. Experiments were conducted at the Sumiyoshi Field of Miyazaki University to confirm the effectiveness of the proposed method.
-
Applying Digital Restoration Techniques in Preservation of Ancient Murals Using Diffusion-Based Inpainting Reviewed International coauthorship International journal
Tint K.K.W., Mie Mie Tin, Thi Thi Zin, Pyke Tin
Lecture Notes in Electrical Engineering 1321 LNEE 398 - 407 2025.2
Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
Nowadays, with the assistance of advanced technology, everything is constantly evolving in every sector, such as transportation, economics, medication, and so on. Also, in the sector of cultural heritage, image inpainting technology has become very popular as an effective way that allows conservators to digitally restore damaged murals without physically altering the original artwork, preserving its integrity and historical authenticity. In this paper, an image inpainting framework is proposed to restore the ancient murals of Myanmar from AD 1800–1900. The framework can be subdivided into crack removal and lacuna removal. The identification of crack and lacuna damage is automatically done with segmentation and image processing methods. Crack damage is reconstructed with pixel neighboring transfer, while lacuna reconstruction is applied with coherent transport and patch-based nearest neighbor similarity color filling methods. The accuracy is tested with the damage ratio analysis, and the experimental result demonstrates that the framework can deliver satisfactory visual results in the reconstruction process of the murals.
-
Automated Cattle Identification via Image-Based Ear Tag Recognition Reviewed International journal
Y. Shimizu, Thi Thi Zin, M. Aikawa, I. Kobayashi
Lecture Notes in Electrical Engineering 1321 LNEE 138 - 147 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
A cattle management system using image processing technology has been proposed to reduce the labor burden of livestock farmers and improve management efficiency. To make cattle management using image processing more efficient, individual identification is necessary. Therefore, we focused on the 10-digit individual identification number given to each cow. In this study, face detection is first performed using YOLOv8, and ear regions are set from the detected face regions. The ear tag region is extracted from the ear region using color information, and the individual identification number on the ear tag is read. The identification result of the ear tag number may not be correctly identified due to the fact that a part of the ear tag is not captured due to the movement of the face or ear. Therefore, we proposed a mechanism to give a confidence score to the character identification results, update the ear tag number identification results, and obtain the ear tag number correctly, and confirmed its effectiveness through a demonstration experiment.
-
Automatic Body Temperature Detection in Calves and Alarm System Using Thermographic Camera Reviewed International journal
Aung Si Thu Moe, Thi Thi Zin, M. Aikawa, I. Kobayashi
Lecture Notes in Electrical Engineering 1321 LNEE 190 - 198 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
The health monitoring of cows is crucial in livestock farming, particularly for calves, which are more susceptible to infectious diseases than adult cattle. This vulnerability is significantly influenced by the maturation of the calf’s immune system, rearing environment, and stress management. Traditional methods of health monitoring require substantial manpower, which can be impractical and inefficient. This paper proposes an automatic body temperature detection system for calves using thermal imaging. Leveraging the capabilities of thermal images, which are widely used in security, medical, and industrial applications, this system aims to identify and monitor the body temperature of calves. By detecting the head and eyes of the calf and extracting temperature data through the Detectron2 object detection method, the system can provide timely notifications to farmers or veterinarians. The overall average detection rate of head and eye regions was 94.5%. This approach enhances the efficiency of livestock health management, reducing the reliance on manual labor and enabling early detection of health issues.
-
Cattle Lameness Classification Using Cattle Back Depth Information Reviewed International journal
San Chain Tun, Pyke Tin, M. Aikawa, I. Kobayashi, Thi Thi Zin
Lecture Notes in Electrical Engineering 1321 LNEE 160 - 170 2025.2
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
The livestock industry plays a crucial role in sustaining agricultural production and rural economies. Monitoring cattle health, however, presents significant challenges on large farms where traditional methods require diagnosing each animal individually. Lameness is a major issue affecting cattle health, leading to decreased production performance on many farms. Timely detection of lameness is essential for providing effective early treatment. In this study, we propose a system using specialized depth cameras to monitor and analyze cattle back information for classifying lameness scores. We employ Detectron2 for cattle detection and segmentation, and the Intersection over Union (IOU) method for tracking, focusing solely on the cattle’s depth region. We extract various features from the cattle’s back depth data and utilize three different machine learning algorithms: K-Nearest Neighbor (KNN), Gradient Boosting, and Extra Trees for lameness score classification. The models KNN, Gradient Boosting, and Extra Trees showed strong training and validation results. Testing showed Extra Trees performing well with 88.2% morning and 89.0% evening accuracy. Our approach demonstrates the potential of depth camera in effectively classifying lameness scores, offering significant implications for livestock health management. This method not only improves the efficiency and accuracy of health monitoring in large-scale farming but also provides a practical solution for early detection and treatment of lameness, thereby enhancing overall farm productivity.
-
Cattle Lameness Detection Using Leg Region Keypoints from a Single RGB Camera Reviewed International journal
Bo Bo Myint, Thi Thi Zin, M. Aikawa, I. Kobayashi, Pyke Tin
Lecture Notes in Electrical Engineering 1321 LNEE 180 - 189 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
The recent rise of machine learning and deep learning has significantly impacted the field of computer vision, particularly in tasks like object detection, object tracking, and classification. This surge in interest has underscored the critical role of feature extraction as a foundational step in these machine learning pipelines. Our research focuses on applying feature extraction techniques to a cattle lameness dataset. We specifically extract features related to the movement of key points on cattle legs across a sequence of video frames. By analyzing the variations in these points, we aim to identify features that can efficiently differentiate between lame and no lame cattle using popular machine learning algorithms. All four classifiers achieved strong testing accuracy above 75%, with SVM excelling at over 84%.
-
Cow’s Back Surface Segmentation of Point-Cloud Image Using PointNet++ for Individual Identification Reviewed International journal
Pyae Phyo Kyaw, Pyke Tin, M. Aikawa, I. Kobayashi, Thi Thi Zin
Lecture Notes in Electrical Engineering 1321 LNEE 199 - 209 2025.2
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
An automatic cow health monitoring and management system cannot function effectively without an identification system in livestock farming. While 2D image-based computer vision currently achieves high accuracy in cow identification, its effectiveness can be significantly decreased by changes in lighting, environmental factors, and other limitations. To address these limitations, an identification system based on point-cloud images will be developed by using a combination of 3D TOF camera and 2D RGB camera. This system includes detection and segmentation stage, feature extraction stage, and identification stage. In this study, I focus on detecting and segmenting of cow back surface region from a point-cloud image using the PointNet+ + algorithm. Two segmentation models are trained and compared based on single-scale grouping (SSG) and multi-scale grouping (MSG) features. The extracted cow back surface region offers a rich set of features valuable for several applications, including individual cow identification, lameness detection, and body condition scoring.
-
Evaluation of Body Condition Score for Walking Dairy Cows Using 3D Camera Reviewed International journal
M. Chikunami, Thi Thi Zin, M. Aikawa, I. Kobayashi
Lecture Notes in Electrical Engineering 1322 LNEE 63 - 72 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
Body Condition Score (BCS) is an important index for assessing body fat accumulation in cattle and plays a crucial role in managing cattle productivity, feeding efficiency, and overall health. Currently, BCS evaluations predominantly rely on visual assessment and palpation by specialized personnel, which is time-consuming and labour-intensive. Consequently, many farms refrain from utilizing BCS for cattle management. Previous studies have focused on BCS evaluation of stationary dairy cows in rotary parlours, but this approach is not feasible for small and medium-sized livestock producers lacking such facilities. To enable BCS management for dairy cows on any farm, we propose a system utilizing image processing technology for evaluating cows while walking. In this study, we capture images using three 3D cameras and construct an estimation model using feature extraction and multiple regression analysis. This model allowed the evaluation of cows with large BCS within an error margin of 0.25. Our findings suggest that this approach can significantly streamline BCS evaluation, making it accessible and practical for a broader range of dairy farms.
-
From Vision to Vocabulary: A Multimodal Approach to Detect and Track Black Cattle Behaviors Reviewed International journal
Su Myat Noe, Thi Thi Zin, Pyke Tin, I. Kobayashi
Lecture Notes in Electrical Engineering 1321 LNEE 171 - 179 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
This paper investigates the potential of recent image-text foundation models for classifying black cattle mounting behavior without fine-tuning. Our approach begins with the detection and tracking of each individual black cattle using deep learning-based, fine-tuned YOLOv9 and Deep OC SORT tracking. Once completed, we employ zero-shot approaches, explicitly utilizing the multi-modal Large Language and Vision Alignment (LLAVA) and Large Language Model Meta AI (LLaMA) models. These models integrate visual and linguistic information seamlessly, enabling us to leverage pre-trained knowledge to analyze and understand black cattle behavior directly from images and accompanying text descriptions. By utilizing zero-shot learning, we can bypass the resource-intensive process of model fine-tuning, making it a highly efficient approach for behavior classification. Our approach highlights the robustness and flexibility of multimodal foundation models like LLAVA and LLaMA in handling complex tasks in the agricultural domain, demonstrating their potential for broader applications without requiring extensive retraining on specific datasets. Through our experiments, we showcase the accuracy and efficiency of this zero-shot multimodal approach, providing valuable insights into black cattle mounting behavior that can enhance livestock management and monitoring practices. We introduced a novel cattle dataset tailored for this purpose. We achieved high detection accuracy with a mAP of 0.9856% using our fine-tuned YOLOv9 model and an average tracking accuracy of 94.79% across five videos. The overall accuracy of our integrated system demonstrates its efficacy in accurately classifying and tracking cattle behaviors, underscoring the potential of zero-shot learning models in precision agriculture.
-
Identification of Rumination Patterns in Cattle Through Optical Flow Analysis and Machine Learning Techniques Reviewed International journal
T. Ishikawa, Thi Thi Zin, M. Aikawa, I. Kobayashi
Lecture Notes in Electrical Engineering 1322 LNEE 134 - 143 2025.2
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：Lecture Notes in Electrical Engineering
Rumination is a critical indicator of a cow’s physiological state, making it a valuable metric for managing cow health and predicting calving. Traditional human observation of rumination behaviour is time consuming and impractical for continuous monitoring, and sensor-based identification can be stressful for the cows due to the need for attachment. To address these challenges, this study introduces a non-contact method for identifying cow rumination. The proposed approach involves capturing video footage of multiple cows from above, recognizing cow regions, and validating the method’s effectiveness through experiments. Specifically, we utilize optical flow and frame-to-frame subtraction methods to extract moving cow regions from the recorded videos. From this data, we derive 11 features and employ a Support Vector Machine (SVM) for classification. Training the SVM with label features resulted in a test data identification accuracy of approximately 60%.