Loading...

Real-time Area Measurement of Agricultural Fields from Aerial Robotic Images

Karimi, Farzam | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57068 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Alasty, Aria
  7. Abstract:
  8. The aim of this research is to measure the area of agricultural lands using realtime aerial images. This project is divided into two parts. In the first part of the project, five segmentation neural networks based on YOLOv8 models were designed. After collecting 734 images of agricultural lands, they were trained, and among them, the nano model was selected with 89% accuracy, 88% recall, and an inference time of 49.8 milliseconds. In the second part of this project, the internal characteristics and distortion elements of the camera used were identified through the calibration process using a chessboard. This process was carried out using 81 images and hadan RMS of 1.178. Also, the elements of the intrinsic matrix were compared with the specifications obtained from the theoretical method, and the error percentage of this comparison was reported as 0.61%. Then, by determining the global coordinates, camera, camera sensor plane, and lens plane, the equations for position transfer from real three-dimensional space to two-dimensional image space were extracted. With constraints applied, the inverse of these equations was also obtained. Afterward, the algorithm for calculating the area of a polygon was selected based on the type of outputs for this problem. Finally, with data collection from the agricultural lands of Fars province, the final test was designed. By extracting the borders of the lands in the images, and having information such as the internal specifications of the camera, the angles of the camera during imaging, and the height, length, and width of the geographical coordinates of the camera, the lands in the image were measured for area. The comparison of the real area of these lands and the obtained area reported an average error of 2.4604% and an average inference time of 0.4456 seconds
  9. Keywords:
  10. Neural Networks ; Agriculture ; Agricultural Drone ; Area Measurement ; Aerial Robotics

 Digital Object List

 Bookmark

...see more