Automatic Longan tree detection from high resolution remote sensing image | |
| Author | Tanawat Chailungka |
| Call Number | AIT Thesis no.RS-15-08 |
| Subject(s) | Remote-sensing images--Thailand Longan--Remote sensing--Thailand |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Remote Sensing and Geographic Information Systems |
| Publisher | Asian Institute of Technology |
| Series Statement | Thesis ; no. RS-15-08 |
| Abstract | Thailand is a major exporter of Longan. Well Longan plantation management is necessary to keep the stable products. Yield factors of plantation Longan depend on weather, soil types, location, tree parameters and farm operation. The number of trees and crown size are another fundamental data which is important for yield estimation. Maejo University is the college which provides agricultural knowledge. In this study, a Longan field of Maejo University is chosen. There are four objectives of this study. The first objective was to detect Longan trees automatically using very high resolution of images. Face detection or Haar classifier which is proposed by Viola and Jones and Color detection were used.The second objective was to check robustness of detection. To estimate the performance of detecting and counting number of Longan trees by Face detection and Color detection, UAV, Google Earth, GeoEye-1 and QuickBird are applied. The result was found that Face detection worked well when pixel size is less than 25 centimeter. It resulted the error of counting trees, 2.61 % and 4.81% in UAV and Google Earth. But GeoEye-1 and QuickBird, the error of detection has rapidly increased to be 48.70 and 98.80 %, when pixel size is bigger than 25 centimeter. In part of Color detection, its performance was related to a pixel size of an image. It provided the error of counting number of Longan trees, 3.60, 11.02, 24.45 and 75.95% in UAV, Google Earth, GeoEye-1 and QuickBird.The third objective was to estimate crown diameter. Theoutputs from Face detection and Color detection was in binary images. The individual crown diameter was estimated from vector file. The last objective was to evaluate the accuracy by field survey. The output from Face detection and Color detection were conducted accuracy assessment by 100 field data which were collected by measuring crown diameter in N-S direction and E-W direction. For the result of UAV image, Face detection and Color detection could detect all 100 tree and provide the overall errors of estimated crown diameter, 0.77 and 0.42 centimeter. In Google Earth, Face detection and Color detection were able to detect 97 and 93 trees, both provided the overall error, 0.85 and 0.48 centimeter for estimating crown diameter. Face detection and Color detection made 1.4 and 0.52 centimeter of the overall errors for crown diameter, they were able to detect 33 and 71 of all 100 tree when using GeoEye-1. In part of QuickBird, the acquisition time is different from previous three images for four years. Face detection and Color detection provide the result of tree detection, 2 and 16 trees, respective. In addition, the overall errors of estimated crown diameter were 0.95 and 0.62 centimeter by Face detection and Color detection. |
| Year | 2015 |
| Corresponding Series Added Entry | Asian Institute of Technology. Thesis ; no. RS-15-08 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Remote Sensing (RS) |
| Chairperson(s) | Nagai, Masahiko |
| Examination Committee(s) | Tripathi, Nitin Kumar;Apichon Witayangkurn |
| Scholarship Donor(s) | Royal Thai Government;AIT Fellowship |
| Degree | Thesis (M. Sc.) - Asian Institute of Technology, 2015 |