Intelligent Agricultural Greenhouse Extraction Method Based on Multifeature Modeling: Fusion of Geometric, Spatial, and Spectral Characteristics

The spatial distribution of agricultural greenhouses plays a crucial role in precision agriculture management and the achievement of sustainable development goals. However, existing methods do not adequately capture the unique characteristics of agricultural greenhouses. This article proposes an intelligent extraction model that effectively addresses the geometric, spatial, and spectral features of agricultural greenhouses in remote sensing images. The model incorporates targeted modeling techniques to account for the greenhouse's large aspect ratio, clustered spatial distribution, and consistent texture and spectral properties. It consists of three main components: a strip-like feature enhancement encoder, a frequency-guided pyramid decoder, and a lightweight foreground enhancement segmentation head. The collaborative interaction of these components enables efficient and accurate segmentation of agricultural greenhouses. Comparative experiments and ablation studies conducted on a self-labeled dataset demonstrate the effectiveness of the proposed method. Experimental results, validated through multiple evaluation metrics, show that the method achieves optimal extraction accuracy with an F1 score of 90.04% and an IoU of 81.89%, while maintaining a strong balance between accuracy and model complexity. This article presents a novel approach to intelligent greenhouse extraction, contributing significantly to the advancement of precision agriculture technologies.