This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint.
Downloads
Authors
Abstract
: Instance segmentation is a novel technique to automatically detect and count the
number of objects from satellite imagery for various applications using deep learning
frameworks such as Mask-RCNN and YOLO. In this paper, we have implemented the
YOLOv5 and YOLOv7 instance segmentation models on high-resolution satellite imagery
(0.31m to 1.74m) of both panchromatic (16-bit PAN) and multi-Spectral (16-bit 9-channels
MS) sensors and evaluated the comparative performance of these models. After training both
the models on 300 epochs, the models showed very good comparative accuracies. YOLOv7
outperformed YOLOv5 on Mean Average Precision (mAP) parameter with of 99.20% mAP
value as compared to YOLOv5's 99.12% mAP value. We have also obtained the model
results on panchromatic and multi-spectral remote sensing data of Indian remote sensing data
over Mumbai, Pune, and Ahmedabad airports with an accuracy of above 94% to segment the
larger aircrafts and above 88% to segment smaller aircrafts
DOI
https://doi.org/10.31223/X5CM5X
Subjects
Engineering
Keywords
: Instance Segmentation, Aircraft Detection, YOLO v7, YOLO v5, Satellite Imagery, Remote Sensing
Dates
Published: 2024-11-11 22:02
Last Updated: 2024-11-12 06:02
License
CC-By Attribution-NonCommercial-NoDerivatives 4.0 International
Additional Metadata
Conflict of interest statement:
none
Data Availability (Reason not available):
Due to nature of the data.
There are no comments or no comments have been made public for this article.