Vision–Language Multimodal State Space Network for UAV-Based Soil Moisture Change Detection
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, cilt.19, ss.35217-35229, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 19
- Basım Tarihi: 2026
- Doi Numarası: 10.1109/jstars.2026.3741947
- Dergi Adı: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Compendex, Geobase, INSPEC, Directory of Open Access Journals, Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Sayfa Sayıları: ss.35217-35229
- Anahtar Kelimeler: Cross-modal fusion, multiscale features, semantics-driven, soil moisture, uncrewed aerial vehicle (UAV) remote sensing
- Manisa Celal Bayar Üniversitesi Adresli: Hayır
Özet
With the increasing application of uncrewed aerial vehicle remote sensing technology in precision agriculture, the detection of changes in soil moisture content using high-resolution imagery faces challenges, such as irregular patch shapes, a lack of semantic information, and difficulties in modeling long-range dependencies. To address these challenges, this article proposes a vision–language multimodal state space network (VLM-SSNet). This method constructs a multiscale deformation-aware feature fusion module to accurately capture irregular soil boundaries and multiscale textures; introduces a vision–language cross-modal interaction module to reduce visual ambiguity using semantic priors; and designs a multipath fusion state space module to achieve efficient global long-range modeling with linear computational complexity. Experimental results demonstrate that VLM-SSNet performs exceptionally well in detecting changes in soil moisture content, achieving an F_{1}-score of 90.92% and an mIoU of 80.14%, representing significant improvements over mainstream methods. Furthermore, the model exhibits excellent generalization capabilities under complex resolution conditions and in the presence of noise interference, providing reliable technical support for the precise monitoring of farmland.