AUTONOMOUS TREE LABELING AND RECOGNITION
Loading...
Date
2023
Authors
Almakhan, Symbat
Journal Title
Journal ISSN
Volume Title
Publisher
School of Engineering and Digital Sciences
Abstract
The process of data labeling is a critical step in the machine learning workflow since
the quality and accuracy of the labeled data can significantly impact the performance
of the trained models. Data labeling can be performed manually or automatically,
depending on the dataset’s complexity and the available resources.Manual data labeling
involves human experts who review and annotate the data based on predefined
guidelines or criteria. While this approach provides high-quality labeled data, it can
be time-consuming, labor-intensive, and costly, particularly for large datasets.
In this research paper, the automatic annotation and classification approach is
evaluated to determine if an autonomous data labeling can be designed for tree classification
problems. The study utilized a dataset of 465 high-resolution images with
dimensions of 6000x4000. K-means clusters the images based on the similarity of
their feature vectors, which are extracted from the images using a feature extractor
model into 4 and 10 classes. The clustered data was then used to train a CNN model,
which was tested on a separate dataset. The results of the predictions were saved in
a new folder with respective class labels and mapped to the original images, with the
location of the cropped images indicated. Mapping the clustered images with their
original images provides a visual understanding of which cluster is characterized by
the features that are mostly present in images belonging to a certain cluster. By
examining the features of each cluster, it is possible to identify the features that are
most commonly associated with a particular image cluster. In this way, it is possible
to identify the cluster that has features that are specific to trees.
Description
Keywords
Type of access: Restricted, data labeling
Citation
Almakhan, S. (2023). Autonomous tree labeling and recognition. School of Engineering and Digital Sciences