The Estimation of Pepper Plant Growth in a Greenhouse Using Regression Based Model

( Thavisack Sihalath ),( Jayanta Kumar Basak ),( Elanchezhian Arulmozhi ),( Ji-hoon Park ),( Fawad Khan ),( Frank Gyan Okyere ),( Yong-jin Lee ),( Jun-hyeon Lee ),( Deog-hyun Lee ),( Hyeon-tae Kim ) 한국농업기계학회 2019 한국농업기계학회 학술발표논문집 Vol.24 No.1

Implementation environmental inside a greenhouse for measurement pepper plant growth and checked relationship between the total leaf area, fresh weight and dry weight and expansion the growth of plant. An experiment was conducted to develop regression models for estimating leaf area, fresh and dry weight from measurement with plant height. The experiment was conducted in a greenhouse with automatic control system, the greenhouse dimension was width 3m, length 4m, height 2.5 m. Inside the greenhouse used sensor located in three different places with three distinct height mean values of temperature and concentration of CO2 were 16.4℃ and 365.2 ppm, with a range of 5.7-30.3℃. Total of 360 pepper cultivated in two beds which training set and validation set. Five models linear regression model; two power polynomial regression model (P. order 2); three power polynomial regression model (P. order 3); four power polynomial regression model (P. order 4) and power were evaluated and compared using the coefficient of determination (R2), Pearson’s correlation (r), root mean square error (RMSE), relative standard error (RSE) and mean absolute percentage error (MAPE). Power regression model involving to estimate in plant height for the expected leaf area (R2>0.96, r>0.98, RMSE<1.12, RSE<0.03, MAPE< 11.27). However P. 2 had more accuracy to calculate the fresh weight (R2 >0.98, r>0.99, RMSE <0.24, RSE<0.03, MAPE< 15.04) and dry weight (R2 >0.97, r>0.98, RMSE<0.03, RSE<0.02, MAPE<11.38) of plant with considering both the fit and degree of adjustment and interpretation of model. This study presents the analysis of various regression models based on data set of pepper plant and creates scope for further species of crops by changing management practices under different environmental condition to understanding of the growing patterns of plants.

Deep Convolutional Neural Network Hyper-Parameters Tuning for Classification Problem

( Thavisack Sihalath ),( Jayanta Kumar Basak ),( Anil Bhujel ),( Byeong Eun Moon ),( Fawad Khan ),( Elanchezhain Arulmozhi ),( Deog Hyun Lee ),( Na Eun Kim ),( Hyeon Tae Kim ) 한국농업기계학회 2020 한국농업기계학회 학술발표논문집 Vol.25 No.1

The hyper-parameter search is the one in the field of deep learning. The hyper-parameters are all the parameters which can be arbitrarily set by the user before starting training. Besides, the image classification is a classical problem of image processing, computer vision and machine learning fields. In this study is presented the experienced the image classification using convolutional neural network namely VGG16 pre-trained model for the observation of hyper-parameters performance and image classification purpose. The dataset is selected from the Gyeongsang National University and Kaggle dataset for the experimentation. The experiments are conducted through different hyper-parameters such as optimizers, batch size and epoch to observe the performance and accuracy classification improvement. It is proposed to build a deep learning model using various optimizers such as SGD, RMSProp, Adam, Adagrad, Adadelta and Ada max to investigate the losses for each optimizer by the loss function (cross-entropy) for evaluation. Besides, the models have evaluated the results by using a confusion matrix to summarize a prediction results on a classification problem by each class. Empirical results demonstrate that the model is run with minimum loss when applied Adagrad optimizer in the case of 16 batch size and 50 epochs. While performing by increasing the number of batch size and epoch, A dam works well in practice among those optimizers. Interestingly, the maximum accuracy is achieved while performing the Adamax optimizer with 120 batch size and 150 epochs. However, the classification performance is also measured by confusion matrix statistical measurement for binary classification test namely accuracy, recall, precision and F1-score. This study, it is created scope for further experimentation on several datasets under different hyper-parameter conditions to find out a suitable of the optimizers for the neural network. Moreover, this study also enhances knowledge and understanding by using different batch size and epoch to improve accuracy for image classification.

Performance Analysis of Different Optimizers, Batch Sizes, and Epochs on Convolutional Neural Network for Image Classification

Thavisack Sihalath,Jayanta Kumar Basak,Anil Bhujel,Elanchezhian Arulmozhi,Byeong-Eun Moon,Na-Eun Kim,Doeg-Hyun Lee,Hyeon-Tae Kim 경상대학교 농업생명과학연구원 2021 농업생명과학연구 Vol.55 No.2

The important thing in the field of deep learning is to find out the appropriate hyper-parameter for image classification. In this study, the main objective is to investigate the performance of various hyper-parameters in a convolutional neural network model based on the image classification problem. The dataset was obtained from the Kaggle dataset. The experiment was conducted through different hyper-parameters. For this proposal, Stochastic Gradient Descent without momentum (SGD), Adaptive Moment Estimation (Adam), Adagrad, Adamax optimizer, and the number of batch sizes (16, 32, 64, 120), and the number of epochs (50, 100, 150) were considered as hyper-parameters to determine the losses and accuracy of a model. In addition, Binary Cross-entropy Loss Function (BCLF) was used for evaluating the performance of a model. In this study, the VGG16 convolutional neural network was used for image classification. Empirical results demonstrated that a model had minimum losses obtain by Adagrad optimizer in the case of 16 batch sizes and 50 epochs. In addition, the SGD with a 32 batch sizes and 150 epochs and the Adam with a 64 batch sizes and 50 epochs had the best performance based on the loss value during the training process. Interestingly, the accuracy was higher while performing the Adagrad and Adamax optimizer with a 120 batch sizes and 150 epochs. In this study, the Adagrad optimizer with a 120 batch sizes and 150 epochs performed slightly better among those optimizers. In addition, an increasing number of epochs can improve the performance of accuracy. It can help to create a broader scope for further experiments on several datasets to perceive the suitable hyper-parameters for the convolutional neural network. Dataset: https://www.kaggle.com/c/dogs-vs-cats/data

Pig Identification Using Deep Convolutional Neural Network Based on Different Age Range

Sihalath Thavisack,Basak Jayanta Kumar,Bhujel Anil,Arulmozhi Elanchezhian,문병은,김현태 한국농업기계학회 2021 바이오시스템공학 Vol.46 No.2

Purpose In this study, the main objectives are to show the performance of deep convolutional neural network in identifying individual pig and investigate the accuracy level of CNN using four datasets made with pig’s face in different growing period. Methods Firstly, the datasets were captured in an experimental pig barn at a different time. Secondly, the datasets were filtered similar images using the structural similarity index measure (SSIM) for data preparation. Finally, face image classification is performed by employing a deep convolutional neural network (DCNN) namely ZFNet model. Results The results have shown that individual pig identification is outperformed while using the same age dataset in training and testing stage with an accuracy rate above 97%. Conclusions The model performed better in a combined dataset which is a combination of all individual data. For future recommendation, it would be beneficial to perform the effectiveness on a large scale of pigs, and a network model should be considered unsupervised learning in case of ageing classification.

Image data acquisition in a real livestock barn for pig identification

( Thavisack Sihalath ),( Jayanta Kumar Basak ),( Anil Bhujel ),( Jae Sung Park ),( Byeong Eun Moon ),( Elanchezhian Arulmozhi ),( Na Eun Kim ),( Hyeon Tae Kim ) 한국농업기계학회 2020 한국농업기계학회 학술발표논문집 Vol.25 No.2

The development of livestock has been increased demand for identification methods for quality control and welfare management through deep learning. Identification of individual pig has become an issue for traceability in a livestock barn. In this paper, show that the performance of a proposed new method demonstrated that a feasibility of identification individual pig to investigate the effects of changeable aspects of the pigs face appearance by ageing during a growing pig from breeding to finishing stage. A deep learning technique convolutional neural network, namely ZFNet is proposed to classify on 5 individual pigs based on face recognition through 4 different datasets for age range classification. The datasets were captured in an experimental livestock barn environment with a different period of time. The results showed that individual pig identification was outperformed while using the same period of time for training and testing dataset with an accuracy rate above of 97% for each class.

Semantic Segmentation of Strawberry Gray Mold Disease using Deep UNet

( Anil Bhujel ),( Jayanta Kumar Basak ),( Elanchezhian Arulmozhi ),( Thavisack Sihalath ),( Fawad Khan ),( Byeong Eun Moon ),( Mustafa Jaihuni ),( Deog Hyun Lee ),( Hyeon Tae Kim ) 한국농업기계학회 2020 한국농업기계학회 학술발표논문집 Vol.25 No.1

Gray mold is a common disease in strawberry, causing a devastating loss. Therefore, it is very important to identify the gray mold disease as early as possible and check the severity of the disease. The gray mold disease on strawberry plants was produced experimentally in a greenhouse located at Gyeongsang National University. The fungus “Botrytis cinerea” is the causal agent of gray mold disease, which was inoculated in the strawberry plants with three different concentrations (103, 105, and 107 MPN per 50 ml). The images of strawberry leaves infected by the disease were captured by the smartphone and pre-processed. In image pre-processing, removed the background of the image, created a mask of the lesion area of each image, and labeled them. A pair of the original image and its annotated mask were manually prepared and split into training and testing sets. A deep learning convolutional neural network-based UNet model was designed and trained by 45 sets of original and annotated pairs of images using heavy data augmentation. The model was trained for 10 epochs with 1000 steps per epoch, and the training accuracy achieved was 98.92%. Then the model was tested by 10 sets of original and annotated images, which gave the highest pixel accuracy of 98.21%. It was also tested by other segmentation metrics like the intersection of union (iu) and dice accuracy. The model provided the highest iu of 87.91% and dice accuracy of 92.91%. From the results, it can be concluded that the deep learning UNet can successfully segment the gray mold disease that occurred in strawberry, helping to identify the disease severity.

The application of Machine learning to estimate microclimate of a model pig barn

( Elanchezhian Arulmozhi ),( Jayanta Kumar Basak ),( Thavisack Sihalath ),( Byeong Eun Moon ),( Hyeon Tae Kim ) 한국농업기계학회 2021 한국농업기계학회 학술발표논문집 Vol.26 No.1

Indoor air temperature (IAT) and indoor relative humidity (IRH) are the prominent microclimatic variables; still, potential contributors that influence the homeostasis of livestock animals reared in closed barns. Further, predicting IAT and IRH encourages farmers to think ahead actively and to prepare the optimum solutions. Therefore, the primary objective of the current literature is to build and investigate extensive performance analysis between popular ML models in practice used for IAT and IRH predictions. Meanwhile, multiple linear regression (MLR), multilayered perceptron (MLP), random forest regression (RFR), decision tree regression (DTR), and support vector regression (SVR) models were utilized for the prediction. This study used accessible factors such as external environmental data to simulate the models. In addition, three different input datasets named S1, S2, and S3 were used to assess the models. From the results, RFR models performed better results in both IAT (R2 = 0.9913; RMSE = 0.476; MAE = 0.3535) and IRH (R2 = 0.9594; RMSE = 2.429; MAE = 1.47) prediction among other models particularly with S3 input datasets. In addition, it has been proven that selecting the right features from the given input data builds supportive conditions under which the expected results are available. Overall, the current study demonstrates a better model among other models to predict IAT and IRH of a naturally ventilated swine building containing animals with fewer input attributes.

The Application of Artificial Neural Networks and Multiple Linear Regression Models to Estimate Body Weight of Yorkshire Pig

( Jayanta Kumar Basak ),( Elanchezhian Arulmozhi ),( Thavisack Sihalath ),( Fawad Khan ),( Anil Bhujel ),( Deog Hyun Lee ),( Hyeon Tae Kim ) 한국농업기계학회 2020 한국농업기계학회 학술발표논문집 Vol.25 No.1

Body weight of pig is an important indicator for their biological functions and the readiness for market. In pig production, it is essential to know pig’s body weight (PBW) for optimized management and feeding practices as well as genetic improvement of animals. Therefore, the experiment was conducted to build-up and evaluate the predictive performance of artificial neural networks (ANNs) and multiple linear regression (MLR) based models for predicting PBW and to analyze the sensitivity of the input variables to identify the influential factors that affect PBW. Two independent experiments were performed in a pig barn located at Gyeongsang National University. The experiment was conducted from 15 September to 15 December in 2018 and 2019 with ten 10-week-old Yorkshire breed pigs. The performance of the models in predicting pig’s body temperature was determined using statistical quality parameters, including coefficient of determination (R²), root mean square error (RMSE) and mean absolute percentage error (MAPE). The result of the study showed that the Feed Forward Back-propagation (FFBP) model with Levenberg-Marquardt training function, tan-sigmoid transfer function and two hidden layers with 16 neurons was selected as the best model. It is also found from the sensitivity analysis, the length of pig (LP) is the most influential factor in predicting PBW in the MLR/ANN models. In conclusion, the input variables may not always be same when associated with PBW. Therefore, further studies on viable alternative breeds with other growth factors under different management conditions might be considered for development of MLR/ANN models.

실험돈사의 실내 공기 온도 예측을 위한 머신러닝 모델

( Elanchezhian Arulmozhi ),( Jayanta Kumar Basak ),( Byeong Eun Moon ),( Thavisack Sihalath ),( Hyeon Tae Kim ) 한국농업기계학회 2020 한국농업기계학회 학술발표논문집 Vol.25 No.1

Intensive swine production requires an environmental control system to guarantee the welfare and profitable. Indoor air temperature (IAT) is a highly influential variable of the indoor microclimate since that has a direct impact on animal growth, feed utilization, and wellbeing. The swine building environment is an uncertain nonlinear system in which classical modeling methods require several input variables to solve it. Therefore, the foremost goal of the current study is to develop a non-complex Machine Learning (ML) model to predict inside AT (IAT) in a naturally ventilated model swine building at Gyeongsang National University, Jinju, South Korea. Six group-housed Yorkshire x Duroc crossbred pigs were utilized for the present trial. A real-time sensor that collects IAT data was assigned as response data sets for the prediction model. A weather station comprising a data logger managed the acquisition of outdoor weather data for every 10 min interval. Three kinds of different input dataset named M1 (all outdoor parameters), M2 (all outdoor parameters including indoor humidity), and M3 (only the selected features) were used for the modeling. The performance of the model was evaluated with the coefficient of determination (R2), the root mean square error (RMSE), and the mean absolute error(MAE). Furthermore, the efficiency of the model was apprised by comparison with a traditional classical model. From the results the ML model along with M3 input performed better (R2 = 0.99; RMSE = 0.76; MAE = 0.61) than the classical model. The performance of the model was validated with a four-fold cross-validation method during the evaluation. The present study developed a simple and powerful ML model to predict the IAT of the swine building, which may integrate into livestock building controller devices through cloud technology in the future.

Predicting indoor air temperature for IoT sensors of a model pig barn

( Elanchezhian Arulmozhi ),( Jayanta Kumar Basak ),( Byeong Eun Moon ),( Thavisack Sihalath ),( Hyeon Tae Kim ) 한국농업기계학회 2020 한국농업기계학회 학술발표논문집 Vol.25 No.2

Intensive swine production requires an environmental control system to guarantee the welfare and profitable. Indoor air temperature (IAT) is a highly influential variable of the indoor microclimate since that has a direct impact on animal growth, feed utilization, and wellbeing. The swine building environment is an uncertain nonlinear system in which classical modeling methods require several input variables to solve it. Therefore, the foremost goal of the current study is to develop a non-complex Machine Learning (ML) model to predict inside AT (IAT) in a naturally ventilated model swine building at Gyeongsang National University, Jinju, South Korea. Six group-housed Yorkshire x Duroc crossbred pigs were utilized for the present trial. A real-time sensor that collects IAT data was assigned as response data sets for the prediction model. A weather station comprising a data logger managed the acquisition of outdoor weather data for every 10 min interval. Three kinds of different input dataset named M1 (all outdoor parameters), M2 (all outdoor parameters including indoor humidity), and M3 (only the selected features) were used for the modeling. The performance of the model was evaluated with the coefficient of determination (R2), the root mean square error (RMSE), and the mean absolute error (MAE). Furthermore, the efficiency of the model was apprised by comparison with a traditional classical model. From the results the ML model along with M3 input performed better (R2 = 0.99; RMSE = 0.76; MAE = 0.61) than the classical model. The performance of the model was validated with a four-fold cross-validation method during the evaluation. The present study developed a simple and powerful ML model to predict the IAT of the swine building, which may integrate into livestock building controller devices through cloud technology in the future.