A carregar...
Publicação
Object detection for single tree species identification with high resolution aerial images
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 3.22 MB | Adobe PDF |
Resumo(s)
Object recognition is one of the computer vision tasks developing rapidly with the invention of Region-based Convolutional Neural Network (RCNN). This thesis contains a study conducted using RCNN base object detection technique to identify palm trees in three datasets having RGB images taken by Unnamed Aerial Vehicles (UAVs). The method was entirely implemented using TensorFlow object detection API to compare the performance of pre-trained faster RCNN object detection models. According to the results, best performance was recorded with the highest overall accuracy of 93.1 ± 4.5 % and the highest speed of 9m 57s from faster RCNN model which was having inceptionv2 as feature extractor. The poorest performance was recorded with the lowest overall accuracy of 65.2 ± 10.9% and the lowest speed of 5h 39m 15s from faster RCNN model which was having inception_resnetv2 as feature extractor.
Descrição
Dissertation submitted in partial fulfilment of the requirements for the degree of Master of Science in Geospatial Technologies
Palavras-chave
Convolutional Neural Network High Resolution Aerial Images Image Classification Object Detection Region-based Convolutional Neural Network Remote Sensing Unnamed Aerial Vehicle