Javascript is required
Search
Volume 5, Issue 1, 2026

Abstract

Full Text|PDF|XML

Large-scale Vision-Language Models (VLMs) like Contrastive Language-Image Pre-training (CLIP) have demonstrated their impressive zero-shot capabilities. However, adapting them to downstream tasks remains challenging, especially under domain shifts where visual features become unreliable. Existing training-free methods, such as Tip-Adapter, rely heavily on visual similarity, which often fails in out-of-distribution (OOD) scenarios. To address this, Decoupled Correction Adapter (DeCo-Adapter), a robust adaptation framework that integrates a Decoupled Knowledge Stream into the visual baseline, is proposed. Specifically, a novel Negative Semantic Suppression mechanism is introduced, leveraging Large Language Models (LLMs) to generate and penalize distractor descriptions. This mechanism effectively corrects visual ambiguities without requiring any training. Extensive experiments on ImageNet-Sketch, ImageNet-V2, and ImageNet-A demonstrate that DeCo-Adapter consistently outperforms state-of-the-art methods. Notably, it achieves a top-1 accuracy of 54.11% on ImageNet-Sketch, surpassing the strong Tip-Adapter baseline by leveraging negative knowledge for error correction.

Abstract

Full Text|PDF|XML
Accelerated urbanization, sustained rural labor migration, and increasing inefficiencies in land management have contributed to the widespread abandonment of cultivated land in China, thereby posing significant challenges to national food security, agricultural sustainability, and rural revitalization. Although intelligent supervision technologies have been increasingly introduced into agricultural governance systems, the heterogeneous requirements of multiple stakeholders have not been systematically incorporated into existing platform design frameworks. To address this gap, a Kano model–based requirements analysis framework was developed and applied to the governance of farmland abandonment in a major agricultural county in Jiangxi Province, China. A mixed-methods approach integrating literature analysis, semi-structured interviews, and questionnaire surveys was adopted to identify, classify, and prioritize the requirements for an intelligent supervision platform. The identified requirements were categorized into four dimensions: must-be requirements (e.g., policy subsidy information and data stability), one-dimensional requirements (e.g., historical data comparison and land transfer information), attractive requirements (e.g., high-precision monitoring and fallow warning), and indifferent requirements (e.g., user operation training and feedback channels). The findings demonstrated that must-be requirements should be prioritized to ensure the operational reliability of the platform, whereas one-dimensional requirements should be continuously strengthened to improve core capabilities. Attractive requirements were found to significantly enhance user experience and should therefore be gradually integrated. In contrast, indifferent requirements should be strategically managed to avoid unnecessary allocation of resources. Empirical evidence for the optimization of intelligent supervision platforms in farmland abandonment governance was provided by this study, while the applicability of the Kano model in public governance technology requirement analysis was further validated. The findings are expected to contribute to the advancement of intelligent, data-driven, and precision-oriented farmland governance systems in China and other developing agricultural regions.
- no more data -