Transformer Encoder and Decoder Method for Forest Estimation and Change Detection

Forest Change Detection (FCD) is difficult component of natural assess monitoring and conservation strategy, enabled informing decision-making. Different techniques using strength of Artificial Intelligence (AI) has implemented to detect and classify modifications in forest cover by Remote Sensing (...

Full description

Saved in:

Bibliographic Details
Published in:	2024 International Conference on Integrated Circuits and Communication Systems (ICICACS) pp. 1 - 4
Main Authors:	Kotin, Kiran Kashinath, Kumar, Sharan, Alabdeli, Haider, Kumar, Gotte Ranjith, A C, Ramachandra
Format:	Conference Proceeding
Language:	English
Published:	IEEE 23-02-2024
Subjects:	Decoding Estimation Forest Change Detection Forestry Generative adversarial networks Integrated circuits Multi Feature Fusion Multi Scale and Multi Decoder Remote Sensing and Transformer Encoder and Decoder Sensors Transformers
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Forest Change Detection (FCD) is difficult component of natural assess monitoring and conservation strategy, enabled informing decision-making. Different techniques using strength of Artificial Intelligence (AI) has implemented to detect and classify modifications in forest cover by Remote Sensing (RS) data. In this research, proposed a Transformer Encoder and Decoder (TED) method for forest estimation and change detection. The Multi Feature Fusion (MFF) is used for effective combine features of various scales. Next, Multilevel decoding named as Multi Scale and Multi Decoder (MSMD) is used at fine the scales much excessively. At last, loss is function is used for minimizing loss of MSMD. The performance of proposed method is analysed with performance measures of Intersection over Union (IoU) and Accuracy. The proposed method attained IoU of 86.23% and 87.45% in forest and background Attained mIoU of 85.21% and 88.92% in forest and background respectively. The proposed method attained accuracy of 92.89% and 92.33% in forest and background. Attained mAcc of 90.54% and 91.48% in forest and background respectively. The proposed method performed effective than previous methods like Multiscale Channelwise Cross Attention Network (MCCANet), UNet and Generative Adversarial Networks (GAN).
DOI:	10.1109/ICICACS60521.2024.10498369