IntentCD: Remote Sensing Change Detection Guided by Implicit User Intent

Remote sensing change detection (CD) is commonly formulated as pixel-level localization of predefined changes or changes specified by explicit semantic categories. However, practical users often express their requirements indirectly through functional, causal, or consequence-oriented descriptions, creating a semantic gap between natural-language intent and the corresponding spatiotemporal change regions. Our introduces implicit user-intent-guided change detection and proposes IntentCD, an intent-aware multimodal framework that translates indirect user requirements into pixel-level change maps. The key idea is to use a latent change-of-interest representation to bridge intent interpretation and temporal change localization. Specifically, bi-temporal remote sensing images and user instructions are jointly encoded into a multimodal representation, while a pretrained large language model is employed to infer implicit user intent from implicit descriptions. The inferred semantic representation is then used to condition a change-map decoder, which integrates high-level semantic features and multi-scale visual features for pixel-level change localization. Experiments are conducted on the BCDD and SECOND datasets. On BCDD, IntentCD achieves an F1 score of 92.1% for building change detection. In addition, implicit-intent annotations derived from multiple land-cover categories in SECOND are used to evaluate the model under indirect semantic descriptions. The results demonstrate that the proposed framework can generate change maps conditioned on both explicit and implicit user requirements while maintaining competitive change detection performance. These findings suggest that incorporating implicit user intent provides a more flexible interface between natural-language requirements and remote sensing change analysis.

Authors

Institutions

Publication Details

Journal
Remote Sensing
Published
2026-11-17
DOI
https://doi.org/10.3390/rs18223951
Primary Topic
Remote-Sensing Image Classification
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

IntentCD: Remote Sensing Change Detection Guided by Implicit User Intent

Zhenyang Huang, Chao Tao, Haifeng Li, Xian Li et al.
3 links
Remote Sensing
Remote-Sensing Image Classification
article

IntentCD: Remote Sensing Change Detection Guided by Implicit User Intent

Zhenyang Huang, Chao Tao, Haifeng Li, Xian Li, Tongxin Wu, Xiao Ma
article en

Abstract

Remote sensing change detection (CD) is commonly formulated as pixel-level localization of predefined changes or changes specified by explicit semantic categories. However, practical users often express their requirements indirectly through functional, causal, or consequence-oriented descriptions, creating a semantic gap between natural-language intent and the corresponding spatiotemporal change regions. Our introduces implicit user-intent-guided change detection and proposes IntentCD, an intent-aware multimodal framework that translates indirect user requirements into pixel-level change maps. The key idea is to use a latent change-of-interest representation to bridge intent interpretation and temporal change localization. Specifically, bi-temporal remote sensing images and user instructions are jointly encoded into a multimodal representation, while a pretrained large language model is employed to infer implicit user intent from implicit descriptions. The inferred semantic representation is then used to condition a change-map decoder, which integrates high-level semantic features and multi-scale visual features for pixel-level change localization. Experiments are conducted on the BCDD and SECOND datasets. On BCDD, IntentCD achieves an F1 score of 92.1% for building change detection. In addition, implicit-intent annotations derived from multiple land-cover categories in SECOND are used to evaluate the model under indirect semantic descriptions. The results demonstrate that the proposed framework can generate change maps conditioned on both explicit and implicit user requirements while maintaining competitive change detection performance. These findings suggest that incorporating implicit user intent provides a more flexible interface between natural-language requirements and remote sensing change analysis.

Remote SensingVol. 18(22)
Central South University (CN), Beijing Academy of Artificial Intelligence (CN)
National Natural Science Foundation of China
Climate action
Remote-Sensing Image Classification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon