Automated Ground Truth Annotation for Forest and Non-Forest Classification in Satellite Remote Sensing Images
Accurate ground truth annotation plays a vital role in training and evaluating deep learning models for forest and non-forest classification tasks. This paper introduces a robust algorithm designed to automate the annotation process, specifically targeting the identification of regions of interest;...
| 總結: | Accurate ground truth annotation plays a vital role in training and evaluating deep learning models for forest and non-forest classification tasks. This paper introduces a robust algorithm designed to automate the annotation process, specifically targeting the identification of regions of interest; either forest or non-forest based on satellite images. The algorithm incorporates advanced image processing techniques, such as Gaussian blur and thresholding, to enhance the accuracy and reliability of the generated ground truth annotations. To determine the optimal sequence of steps, the results from a pilot test revealed that applying Gaussian blur before grayscale and thresholding operation yielded the best performance. The proposed algorithm presents a systematic and efficient approach to generate the ground truth annotations, effectively reducing manual effort and minimizing subjectivity among human annotators. Evaluation against a set of referenced ground truth dataset showcases the algorithm's effectiveness, with an impressive SSIM value of 0.874640, thus, accurately identifying the forest and non-forest regions. The study contributes to the field of forest monitoring by offering a valuable tool to researchers engaged in land cover analysis using satellite imagery. Moreover, it paves the way for future advancements in automated annotation techniques. The study uniquely emphasizes the aim of minimizing biases and uncertainties, which stands as a novel and pivotal contribution. The study underscores the significance of harnessing cutting-edge technology to foster efficiency, dependability, and precision in analyzing complex forest ecosystems. © 2023 IEEE. |
|---|---|
| ISSN: | |
| DOI: | 10.1109/AiDAS60501.2023.10284683 |