Palm Oil Tree Canopy Identification Using Deep Learning Approach (Case Study: Tanjung Gusta District, North Sumatera) DOI Open Access

Nurul Fitri Alya,

Hepi Hapsari Handayani, Reza Fuad Rachmadi

et al.

IOP Conference Series Earth and Environmental Science, Journal Year: 2024, Volume and Issue: 1418(1), P. 012011 - 012011

Published: Dec. 1, 2024

Abstract The palm oil plantation industry in Indonesia has growing rapidly as demand for increases globally. This needs to be supported by technological innovation increase production. One of them is integrate the power artificial intelligence technology. research aims develop a robust and accurate method segmenting trees areas. Leveraging deep learning algorithms techniques, explores potential SAM accurately delineating individual derived from aerial imagery data. study also involves development comprehensive versatile labelled dataset support training validation models tree counting segmentation. performance proposed approach evaluated discussed critically. demonstrates large-scale mapping author hopes that result analysis this will give insight improvement detecting using automatic method.

Language: Английский

A Domain-Adaptive Segmentation Method Based on Segment Anything Model for Mechanical Assembly DOI
Jinlei Wang, Chengjun Chen, Chenggang Dai

et al.

Measurement, Journal Year: 2024, Volume and Issue: 235, P. 114901 - 114901

Published: May 12, 2024

Language: Английский

Citations

4
SAM-Enhanced Cross-Domain Framework for Semantic Segmentation: Addressing Edge Detection and Minor Class Recognition DOI Open Access
Qian Wan, Haoxiang Su,

Xianyun Liu

et al.

Processes, Journal Year: 2025, Volume and Issue: 13(3), P. 736 - 736

Published: March 3, 2025

Unsupervised domain adaptation (UDA) enables training a model on labeled source data to perform well in target without supervision, which is especially valuable vision-based semantic segmentation. However, existing UDA methods often struggle with accurate labeling at object boundaries and recognizing minor categories the domain. This paper introduces novel framework—SamDA—that incorporates Segment Anything Model (SAM), large-scale foundational vision model, as mask generator enhance edge segmentation performance. The framework comprises three core modules: cross-domain image mixing module, self-training module teacher–student network, exponential moving average (EMA). It also includes finetuning that leverages SAM-generated masks for pseudo-label matching. Evaluations GTA5 Cityscapes datasets demonstrate SamDA achieves mean IoU (mIoU) of 75.2, surpassing state-of-the-art such MIC-DAFormer by 1.0 mIoU outperforming all ResNet-based approaches least 15 mIoU. Moreover, significantly enhances small objects like bicycles, riders, fences, with, respective, improvements 4.5, 5.2, 3.8 compared baseline models.

Language: Английский

Citations

0
Adapting SAM for Visible-Light Pupil Segmentation Baseline DOI Open Access
O. O. Mil’man, Dovi Yellin, Yehudit Aperstein

et al.

Electronics, Journal Year: 2025, Volume and Issue: 14(9), P. 1850 - 1850

Published: May 1, 2025

Pupil segmentation in visible-light (RGB) images presents unique challenges due to variable lighting conditions, diverse eye colors, and poor contrast between iris pupil, particularly individuals with dark irises. While near-infrared (NIR) imaging has been the traditional solution for eye-tracking systems, accessibility practicality of RGB-based solutions make them attractive widespread adoption consumer devices. This paper a baseline RGB pupil by adapting Segment Anything Model (SAM). We introduce multi-stage fine-tuning approach that leverages SAM’s exceptional generalization capabilities, further enhancing its elemental capacity accurate segmentation. The staged consists SAM-BaseIris enhanced detection, SAM-RefinedIris improving automated bounding box prompts, SAM-RefinedPupil precise Our method was evaluated on three standard datasets: UBIRIS.v2, I-Social DB, MICHE-I. results demonstrate robust performance across conditions colors. achieves near SOTA attains mean mIOU DICE scores 79.37 87.79, respectively, datasets. work establishes strong foundation systems demonstrates potential models specialized medical tasks.

Language: Английский

Citations

0
CoDA: Instructive Chain-of-Domain Adaptation with Severity-Aware Visual Prompt Tuning DOI
Ziyang Gong, Fuhao Li, Yupeng Deng

et al.

Lecture notes in computer science, Journal Year: 2024, Volume and Issue: unknown, P. 130 - 148

Published: Jan. 1, 2024

Language: Английский

Citations

2
Palm Oil Tree Canopy Identification Using Deep Learning Approach (Case Study: Tanjung Gusta District, North Sumatera) DOI Open Access

Nurul Fitri Alya,

Hepi Hapsari Handayani, Reza Fuad Rachmadi

et al.

IOP Conference Series Earth and Environmental Science, Journal Year: 2024, Volume and Issue: 1418(1), P. 012011 - 012011

Published: Dec. 1, 2024

Abstract The palm oil plantation industry in Indonesia has growing rapidly as demand for increases globally. This needs to be supported by technological innovation increase production. One of them is integrate the power artificial intelligence technology. research aims develop a robust and accurate method segmenting trees areas. Leveraging deep learning algorithms techniques, explores potential SAM accurately delineating individual derived from aerial imagery data. study also involves development comprehensive versatile labelled dataset support training validation models tree counting segmentation. performance proposed approach evaluated discussed critically. demonstrates large-scale mapping author hopes that result analysis this will give insight improvement detecting using automatic method.

Language: Английский

Citations

0