Papers - THI THI ZIN
-
PREDICTING DAIRY COW CALVING TIME USING MARKOV MONTE CARLO SIMULATION AND NAÏVE BAYES CLASSIFIER Reviewed
Thi Thi Zin, Swe Zar Maw, Pyke Tin, Y. Horii, H. Hama
ICIC Express Letters, Part B: Applications 14 ( 8 ) 877 - 888 2023.8
Authorship：Lead author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters, Part B: Applications
This study proposes an approach that utilizes both Markov Monte Carlo simulation and the Bayesian method to predict the time of calving events in dairy cows. Continuous video surveillance was conducted on 30 individual dairy cows 24 hours before calving. Behaviors such as lying, transition from lying to standing, standing, and transition from standing to lying were annotated for each cow between 72 to 168 hours prior to calving. The probabilities for each behavior were derived and used in Markov Monte Carlo simulations to generate behavior patterns of each cow before calving. Three types of datasets, actual, simulated, and a mixture of the two, were investigated using Naïve Bayes Classifier for prediction. The experimental results showed that the hybrid approach accurately classified the calving event, cent by cent. This approach can assist farmers and veterinarians in making informed decisions and taking appropriate actions before the calving event, ultimately improving the health and welfare of dairy cows.
-
Quantifying Body Movement Variability for Cattle Lameness Detection Using Imaging Reviewed International journal
Thi Thi Zin and Pyke Tin
ICIC Express Letters, Part B: Applications 14 ( 7 ) 735 - 74 2023.7
Authorship：Lead author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters, Part B: Applications
This paper proposes a framework for detecting cattle lameness by quantifying the variability of body movement using depth imaging data collected while cows walk from the milking center to the resting area. The framework identifies critical factors that determine lameness scores based on the root mean square successive differences, various types of information entropies, and geometric measures of the collected depth data. To analyze lameness status, we developed an operational simulation model that combines Monte Carlo simulation with popular probability distribution functions such as uniform, normal, Poisson, and Gamma distributions. The simulation results suggest that detection performance and the characteristics of lame and non-lame cows significantly affect body movement variability. By using real-life data, we aim to validate this conjecture in future work.
-
Artificial Intelligence Fusion in Digital Transformation Techniques for Lameness Detection in Dairy Cattle Reviewed
Thi Thi Zin, Ye Htet, San Chain Tun and Pyke Tin
International Journal of Biomedical Soft Computing and Human Sciences (IJBSCHS) 28 ( 1 ) 1 - 8 2023.7
Authorship：Lead author, Corresponding author Language：English Publishing type：Research paper (scientific journal)
-
Facial Region Analysis for Individual Identification of Cows and Feeding Time Estimation Reviewed International journal
Yusei Kawagoe, Ikuo Kobayashi, Thi Thi Zin
Agriculture (Switzerland) 13 ( 5 ) 2023.5
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (scientific journal)
With the increasing number of cows per farmer in Japan, an automatic cow monitoring system is being introduced. One important aspect of such a system is the ability to identify individual cows and estimate their feeding time. In this study, we propose a method for achieving this goal through facial region analysis. We used a YOLO detector to extract the cow head region from video images captured during feeding with the head region cropped as a face region image. The face region image was used for cow identification and transfer learning was employed for identification. In the context of cow identification, transfer learning can be used to train a pre-existing deep neural network to recognize individual cows based on their unique physical characteristics, such as their head shape, markings, or ear tags. To estimate the time of feeding, we divided the feeding area into vertical strips for each cow and established a horizontal line just above the feeding materials to determine whether a cow was feeding or not by using Hough transform techniques. We tested our method using real-life data from a large farm, and the experimental results showed promise in achieving our objectives. This approach has the potential to diagnose diseases and movement disorders in cows and could provide valuable insights for farmers.
-
Facial Region Analysis for Individual Identification of Cows and Feeding Time Estimation Reviewed International journal
Y. Kawagoe, I. Kobayashi, Thi Thi Zin
Agriculture Switzerland 13 ( 5 ) 2023.5
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：Agriculture Switzerland
With the increasing number of cows per farmer in Japan, an automatic cow monitoring system is being introduced. One important aspect of such a system is the ability to identify individual cows and estimate their feeding time. In this study, we propose a method for achieving this goal through facial region analysis. We used a YOLO detector to extract the cow head region from video images captured during feeding with the head region cropped as a face region image. The face region image was used for cow identification and transfer learning was employed for identification. In the context of cow identification, transfer learning can be used to train a pre-existing deep neural network to recognize individual cows based on their unique physical characteristics, such as their head shape, markings, or ear tags. To estimate the time of feeding, we divided the feeding area into vertical strips for each cow and established a horizontal line just above the feeding materials to determine whether a cow was feeding or not by using Hough transform techniques. We tested our method using real-life data from a large farm, and the experimental results showed promise in achieving our objectives. This approach has the potential to diagnose diseases and movement disorders in cows and could provide valuable insights for farmers.
-
Validation and Discussion of Severity Evaluation and Disease Classification Using Tremor Video Reviewed International journal
T. Hayashida, T. Sugiyama, K. Sakai, N. Ishii, H. Mochizuki, Thi Thi Zin
Electronics (Switzerland) 12 ( 7 ) 2023.4
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：Electronics (Switzerland)
A tremor is a significant symptom of Parkinson’s disease, but it can also be a characteristic of essential tremor, thereby hampering even specialists’ ability to differentiate between the two. This study proposes a system that leverages a single RGB camera to evaluate tremor severity and support the differential diagnosis of Parkinson’s disease and essential tremor. The system captures motor symptoms, performs time–frequency analysis using wavelet transforms, and classifies severity and disease using linear classification models. The results showed an accuracy rate of 0.56 for disease classification and 0.50 for severity classification (with an acceptable accuracy rate of 0.96). The analysis indicated that there was a low level of correlation between disease and each feature and a moderate correlation (about 0.6) between severity and each feature. Based on these results, this study recommends classifying severity with a linear model and disease with a nonlinear model to obtain improved accuracy.
-
Comparing State-of-the-Art Deep Learning Algorithms for the Automated Detection and Tracking of Black Cattle Reviewed International journal
Su Myat Noe, Thi Thi Zin, Pyke Tin, I. Kobayashi
Sensors (Basel) 23 ( 1 ) 2023.1
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal)
-
Cattle Face Detection with Ear Tags Using YOLOv5 Model Reviewed International journal
Wai Hnin Eaindrar Mg, Thi Thi Zin
ICIC Express Letters, Part B: Applications 14 ( 1 ) 65 - 72 2023.1
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters, Part B: Applications
With the increasing population’s need for meat and requirements for high food quality, the livestock industry is developing from small-scale and subsistence farming towards intensive and specialized grazing. Cattle monitoring and management system is crucial to be registered for breeding association, food quality tracing, disease prevention and control and fake insurance claims. This research presents cattle face detection with their ear tags’ names by applying light-weight YOLOv5 (You Only Look Once) model. This research is intent to the farmers who can not only monitor and manage the cattle conditions at the farm. The proposed system was trained to get the best accuracy model. The accuracy of the proposed model achieves up to 99.4% for four surveillance cameras.
-
Predicting Calving Time of Dairy Cows by Autoregressive Integrated Moving Average (ARIMA) Model and Exponential Smoothing Model Reviewed International journal
Tunn Cho Lwin, Thi Thi Zin and Pyke Tin
ICIC Express Letters, Part B: Applications 14 ( 1 ) 73 - 79 2023.1
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters, Part B: Applications
Calving time prediction is an important factor in dairy farming. The careful monitoring of cows can help to decrease the loss of calf rates during the calving time; moreover, to know the exact time of birth is crucial to make sure timely assistance. However, direct visual observation is time-wasting for observers, and the continuous presence of observers during calving time may disturb cows. In this study, the recording from video cameras and counting the number of standing-to-lying and lying-to-standing transitions of 25 cows from a large farm at Oita prefecture, Japan, before 72 hours of calving time are applied. To be specific, we model the number of changes in behaviors of standing and lying as a time series in hourly basis. The time series approaches, namely the exponential distribution probability, autoregressive integrated moving average (ARIMA) model, and double exponential smoothing (DES) model, are applied to predicting the calving time and the root mean square error (RMSE) is used to check the accuracy and error value of the experiment. By investigating the changes in behavior patterns a few days before the calving events, the proposed method can predict accurately the time of occurrence of calving events by the developed ARIMA (2,0,1) model. Therefore, the developed model can be used to estimate the calving time which has significantly positive impact for livestock specialists.
-
Comparing State-of-the-Art Deep Learning Algorithms for the Automated Detection and Tracking of Black Cattle Reviewed International journal
Su Myat Noe, Thi Thi Zin, Pyke Tin, I. Kobayashi
Sensors 23 ( 1 ) 532 2023.1
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：Sensors
Effective livestock management is critical for cattle farms in today’s competitive era of smart modern farming. To ensure farm management solutions are efficient, affordable, and scalable, the manual identification and detection of cattle are not feasible in today’s farming systems. Fortunately, automatic tracking and identification systems have greatly improved in recent years. Moreover, correctly identifying individual cows is an integral part of predicting behavior during estrus. By doing so, we can monitor a cow’s behavior, and pinpoint the right time for artificial insemination. However, most previous techniques have relied on direct observation, increasing the human workload. To overcome this problem, this paper proposes the use of state-of-the-art deep learning-based Multi-Object Tracking (MOT) algorithms for a complete system that can automatically and continuously detect and track cattle using an RGB camera. This study compares state-of-the-art MOTs, such as Deep-SORT, Strong-SORT, and customized light-weight tracking algorithms. To improve the tracking accuracy of these deep learning methods, this paper presents an enhanced re-identification approach for a black cattle dataset in Strong-SORT. For evaluating MOT by detection, the system used the YOLO v5 and v7, as a comparison with the instance segmentation model Detectron-2, to detect and classify the cattle. The high cattle-tracking accuracy with a Multi-Object Tracking Accuracy (MOTA) was 96.88%. Using these methods, the findings demonstrate a highly accurate and robust cattle tracking system, which can be applied to innovative monitoring systems for agricultural applications. The effectiveness and efficiency of the proposed system were demonstrated by analyzing a sample of video footage. The proposed method was developed to balance the trade-off between costs and management, thereby improving the productivity and profitability of dairy farms; however, this method can be adapted to other domestic species.
DOI： 10.3390/s23010532
-
Black Cow Tracking by Using Deep Learning-Based Algorithms Reviewed International journal
Cho Cho Aye, Thi Thi Zin, I. Kobayashi
ICIC Express Letters, Part B: Applications 13 ( 12 ) 1313 - 1319 2022.12
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters, Part B: Applications
Raising livestock is essential in the farming industry to meet the consumer’s requirement. Livestock monitoring system is useful to monitor their health without need-ing too much manpower. Thus, livestock tracking becomes one of the vital parts of livestock monitoring system. The objective of this proposed system is to track black cows based on detected features. Here, the YOLOv5 (You Only Look Once) model was used in detection phase to detect cow regions and Deep SORT (Simple Online Real-time Tracking) was applied to tracking the target cows in every consecutive frame. In Deep SORT, it includes appearance feature model to recognize cow’s visual appearance such as shape, size, and pose. The proposed system was best trained by adopting transfer learning method. The detection model achieves an accuracy of 0.995 mAP@0.5 whereas the tracking model gets the performance results in video-1 and video-2 with 99.4% and 98.9%, respectively.
-
Cattle Pose Classification System Using DeepLabCut and SVM Model Reviewed International journal
May Phyu Khin, Thi Thi Zin, Cho Cho Mar, Pyke Tin, Y. Horii
GCCE 2022 - 2022 IEEE 11th Global Conference on Consumer Electronics 494 - 495 2022.10
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：GCCE 2022 - 2022 IEEE 11th Global Conference on Consumer Electronics
This paper proposes the cattle pose classification system by using video-based tracking result. The proposed system is composed of two processes namely feature extraction and classification. In the feature extraction, we employ DeepLabCut network to obtain location feature points which are to be combined with cattle bounding box region values. For classification process, the SVM (Support Vector Machine) classifier will be used. To confirm the proposed method, we tested some experimental results by using the video sequences taken in some real-life dairy farms and classify six poses such as 'standing', 'sitting', 'eating', 'drinking', 'sitting with leg extend' and 'tail raised'. We got average 88.75% accuracy for all poses.
-
Video-based Automatic Cattle Identification System Reviewed International journal
Su Larb Mon, Thi Thi Zin, Pyke Tin, I. Kobayashi
GCCE 2022 - 2022 IEEE 11th Global Conference on Consumer Electronics 490 - 491 2022.10
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：GCCE 2022 - 2022 IEEE 11th Global Conference on Consumer Electronics
In this paper, we propose a method to identify the cattle by using video sequences. In order to do so, we first collect 360-degree top-view video sequences to form dataset. The proposed system is composed of two parts: cattle detection and cattle identification. In the detection process, we utilize YOLOv5(Y ou Only Look Once) model to detect the cattle region in the lane. In this stage, cattle's location and region information are extracted and the cropped images of detected cattle regions are saved for the next stage. We then apply Convolutional Neural Network model (VGG16) to extract the features which will be used to identify individual cattle. For the classification, the proposed system used two supervised machine learning methods, Random Forest and SVM (Support Vector Machine). The accuracy of Random Forest is 98.5% and the accuracy of SVM is 99.6%. After comparing the accuracy rate of two methods, SVM get the better accuracy result. The proposed system achieved the accuracy of over 90% for both cattle detection and identification.
-
Cow Lameness Detection Using Depth Image Analysis Reviewed International journal
San Chain Tun, Thi Thi Zin, Pyke Tin, I. Kobayashi
GCCE 2022 - 2022 IEEE 11th Global Conference on Consumer Electronics 492 - 493 2022.10
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：GCCE 2022 - 2022 IEEE 11th Global Conference on Consumer Electronics
In this paper, we introduce to detect cow lameness by using a depth video camera from the top view position. To classify cow lameness, we first extract the sequences of depth value of the cow body region and the maximum value of the back cow area. Then, we will find the average from the maximum height values of the cow backbone area. By using the average values as a feature vector, we classify the cow lameness with the aid of the Support Vector Machine (SVM). To confirm, we perform some experiments by using depth camera images on a real-life dairy farm. The experimental result shows that our proposed method is promising.
-
Color Space Conversion Technique for Cattle Region Extraction with Application to Estrus Detection Reviewed International journal
Y. Hashimoto, H. Hama, Thi Thi Zin
ICIC Express Letters 16 ( 10 ) 1095 - 1100 2022.10
Authorship：Last author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters
In modern dairy and beef farming with no exception in Japanese livestock industry, an accurate and timely estrus (heat) detection is an important and key factor in efficient and profitable reproductive management performance of cattle herd. Failure in heat detection is costly to the producer and it is considered the critical component of reproductive management. Among many estrus behaviors, visual postures of an individual cow can be successfully recognized and utilized for heat detection. In this aspect, to achieve precise identification and to obtain individual cattle information, extracting cattle region from its background is the fundamental and important step. In general, the inter-frame difference and the background subtraction are widely known as methods to detect moving objects in video images. However, these conventional methods do not work well in Japanese black cattle environments, due to their slow movements. At the same time, since the skin is similar to soil in color, region extraction is not so easy, even if background subtraction is used. Therefore, in this paper, we propose a new method for extracting cattle regions using color space conversion. The proposed method is able to automatically extract cattle regions and tracked cattle from change of the gravity center of the extracted cattle regions. Experimental results show that our approach is effective and promising with high accuracy.
-
HMM-Based Action Recognition System for Elderly Healthcare by Colorizing Depth Map Reviewed International journal
Ye Htet, Thi Thi Zin, Pyke Tin, H. Tamura, K. Kondo, E. Chosa
International Journal of Environmental Research and Public Health 19 ( 19 ) 2022.10
Authorship：Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：International Journal of Environmental Research and Public Health
Addressing the problems facing the elderly, whether living independently or in managed care facilities, is considered one of the most important applications for action recognition research. However, existing systems are not ready for automation, or for effective use in continuous operation. Therefore, we have developed theoretical and practical foundations for a new real-time action recognition system. This system is based on Hidden Markov Model (HMM) along with colorizing depth maps. The use of depth cameras provides privacy protection. Colorizing depth images in the hue color space enables compressing and visualizing depth data, and detecting persons. The specific detector used for person detection is You Look Only Once (YOLOv5). Appearance and motion features are extracted from depth map sequences and are represented with a Histogram of Oriented Gradients (HOG). These HOG feature vectors are transformed as the observation sequences and then fed into the HMM. Finally, the Viterbi Algorithm is applied to recognize the sequential actions. This system has been tested on real-world data featuring three participants in a care center. We tried out three combinations of HMM with classification algorithms and found that a fusion with Support Vector Machine (SVM) had the best average results, achieving an accuracy rate (84.04%).
-
A Special Type of Markov Branching Process Model for the Novel Coronavirus (Covid-19) Outbreak Reviewed International journal
Thi Thi Zin, Pyke Tin, H. Hama
International Journal of Innovative Computing, Information and Control 18 ( 4 ) 1339 - 1346 2022.8
Authorship：Lead author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：International Journal of Innovative Computing, Information and Control
Mathematical modeling has been an important tool to estimate key factors of the transmission and investigate the dynamical system of evolutionary nature in epidemics. More precisely, the outbreaks of the virus or epidemiology is generally considered as an application of branching process. Therefore, in this paper, we propose a special type of Markov branching process model to examine and explore some problems of the novel Coronavirus (COVID-19) infectious disease with the aims of reducing the effective reproduction number of an infection below unity. Since the COVID-19 has been recognized as a global pandemic, we have assessed a big amount of data such as hourly contagious, hospitalized patients, recovered and deaths. However, these data are necessary to be further processed to produce useful information for people and authorities when they make an efficient and optimal decisions. In such a decision-making process, we establish a special type of Gama Markov branching process model which has been successfully applied in other research areas such as queueing and waiting lines problems, stochastic reservoir problems, inventory controls and operation research. Specifically, we develop a three parameter Gama Markov branching process model that is structured in two parts, initial and latter transmission stages, so as to provide a comprehensive view of the virus spread through basic and effective reproduction numbers respectively, along with the probability of an outbreak sizes and duration. As an illustration, we have performed some simulations based on the daily data appearing on WHO dashboard in order to analyze the first semiannual spread of the ongoing Coronavirus pandemic in the region of Myanmar. The results show that the proposed model can be utilized for the real-life applications.
-
An Intelligent Vision-Based Approach for Work Group Identification through Helmet Detection Reviewed International journal
S. Inoue, I. Hidaka, Thi Thi Zin
ICIC Express Letters, Part B: Applications 13 ( 5 ) 511 - 517 2022.5
Authorship：Last author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters, Part B: Applications
Helmets are essential equipment to protect workers from danger during inspection and operation in almost all industries. There is a growing necessity of developing innovative methods to automatically monitor safety and work group identification at industry work sites. With the rapid development of artificial intelligence (AI) based image recognition technologies, computer vision-based inspections have been one of the most important industrial application areas for automation. Thus, in this paper, we propose an intelligent computer vision approach for work group identification through helmet detection by analyzing images collected from 4K camera installed overhead at work site. For this purpose, we attach a marker on the top of the worker’s helmet to detect the helmet and identify the work group. This approach is tested on our data set through simulated experiments and the average accuracy of helmet detection is 92.9%.
-
A Deep Learning Method of Edge-Based Cow Region Detection and Multiple Linear Classification Reviewed International journal
Thi Thi Zin, Saw Zay Maung Maung, Pyke Tin
ICIC Express Letters, Part B: Applications 13 ( 4 ) 405 - 412 2022.4
Authorship：Lead author, Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：ICIC Express Letters, Part B: Applications
In this paper we propose a deep learning method of cow region detection and multiple linear model for classifying behaviors of pregnant cows prior to the occurrence of calving events. Dairy farm management experts and farmers have been well recognized that video monitoring to an individual cow plays an important and significant role in production and re-production processes. To be specific we can learn through video monitoring cow’s health conditions, body conditions even the occurrences of calving difficulties in times. Moreover, due to the advances in the latest computer vision and image processing algorithms, it is possible to develop a camera system that automatically detects the cow’s conditions at a low cost. The fundamental and foremost important step in the proposed system is to detect and segment the cow regions in the video sequences. After the detection process we performed a multiple linear model to classify some behaviors of the detected cows. In particular we consider four states of cow behaviors such as lying state, transition on state from lying to standing, standing state and transition state of standing to lying which are important in studying dairy cow management systems. In order to confirm the validity of our proposed method some experiments are carried out by establishing the video monitoring cameras at the maternity pens of a large dairy farm in Japan. The experimental results show that the proposed method gives an impression of promising with high accuracy.
-
Individual Identification of Cow Using Image Processing Techniques Reviewed
Y. Kawagoe, Thi Thi Zin, I. Kobayashi
2022 IEEE 4th Global Conference on Life Sciences and Technologies (LifeTech) 570 - 571 2022.3
Authorship：Corresponding author Language：English Publishing type：Research paper (international conference proceedings) Publisher：LifeTech 2022 - 2022 IEEE 4th Global Conference on Life Sciences and Technologies
Cow identification has become important in recent years due to outbreaks of diseases such as bovine spongiform encephalopathy. Conventional identification methods available are not efficient, affordable, non-invasive, and cost- effective. Among them, some methods are based on biological markers, such as muzzle point matching and facial recognition. Facial images are the most common biometric characteristics used by humans to identify individuals, and they have received much attention. In this study, we used RGB camera to identify individual cow by their faces and confirmed the effectiveness of this method.