"迁移学习混合专家分类模型：解决高分辨遥感影像场景分类精度低下问题"

The document "Transfer Learning Mixed Expert Classification Model for High-Resolution Remote Sensing Image Scene Classification" addresses the issue of low classification accuracy in small-sample remote sensing image scene datasets due to the diversity and complexity of surface objects. The proposed Transfer Learning Mixed Expert (TLMoE) classification model leverages both global information from fully connected layer features and local detail information from convolutional layer features to achieve more accurate scene classification. The model includes a prediction channel based on fully connected layer features, which utilizes global information to preliminarily classify all scene categories. It also incorporates expert channels for training dedicated expert networks for each scene category, extracting key local information embedded in convolutional layer features to differentiate subtle differences between similar scenes and achieve fine-grained recognition. The combination of prediction weights allows for classification that takes into account both global and local scene differences. Experimental results on small-sample datasets demonstrate that the proposed method effectively recognizes confusing scenes and achieves good classification performance.