УДК 004.85
M. Zakharova
Zaporizhzhia National Technical University
Zaporizhzhia, Ukraine
E-mail:
DEEP LEARNING FOR AUTOMATED OBJECT DETECTION IN AERIAL IMAGERY
M. Zakharova. Deep Learning for Automated Object Detection in Aerial Imagery. The goal of the work is to propose deep learning techniques for coconut trees detection in aerial imagery. The research was made in the TensorFlow framework. The classification was conducted using convolution neural network model obtained through transfer learning algorithms by retraining the pre-trained Inception V3 model. The image dataset for detection was collected through the sliding window technique from the aerial image of the coconut trees plantation area. The experiments with a different amount of training samples and different sliding window techniques were performed to collect the data for the classification. The research work showed deep learning techniques might appear to be promising for the coconut trees detection and counting.
Keywords: deep learning, object detection, coconut trees, convolution neural network (CNN), transfer learning, pre-trained model, TensorFlow framework, sliding window, overlapping.
| Файл | Розмір файла: |
|---|---|
 Повний текст роботи | 363 kB |
