Time-Lapse Of Plant Movement Classification

• Main Author: Azreezul, Azrul Zhafran
• Format: Monograph
• Language: English
• Published: Universiti Sains Malaysia 2018
• Subjects: T Technology; TJ Mechanical engineering and machinery
• Online Access: http://eprints.usm.my/54388/1/Time-Lapse%20Of%20Plant%20Movement%20Classification_Azrul%20Zhafran%20Azreezul_M4_2018.pdf; http://eprints.usm.my/54388/

Plant motions are commonly explored from perspectives of its responses to wind and water. As plant motion is too slow to be observed quickly, the time-lapse technology offers the solution. Previous studies have explored the movement of the plants by applying the tree modelling and botanical simulation. The data mining concept to study plant movements patterns is not adopted yet. However, not many papers reported the plant movement patterns in response to external perturbations such as wind, heat, light and water. Therefore, the goals of this study are to classify the plants responses by perturbation: wind and water, differentiate the classes by plant type in response to wind and water perturbations and compare the branches movement patterns towards wind or water. An experiment was conducted on time-lapse captures on five types of potted plants in response to wind and water. Six markers were placed on identified locations of tree branches (top, middle and bottom) to enable the motion tracking. The videos were translated into numeric data for which the changes in patterns of plants biomotion will be quantitatively analysed using data mining approach. Stages involved include (i) data preprocessing, (ii) classification (iii) knowledge discovery. Data preprocessing techniques include normalize, standardize and remove potential outlier and extreme value. The plants motion are grouped into its attribute classes: perturbation and plant type based on Decision Tree and Lazy classifiers built-in Weka tool. Further analysis was performed to examine the type of plant and location of markers that result in misclassifications. Findings from this study show that 91.1745% classification accuracies were retrieved on J48 classifier for perturbation while 78.8992% for type of plants.