An Effective Task Sampling Strategy Based on Category Generation for Fine-Grained Few-Shot Object Recognition

The recognition of fine-grained objects is crucial for future remote sensing applications, but this task faced with the few-shot problem due to limited labeled data. In addition, existing learning methods do not consider unique characteristics objects, i.e., complex backgrounds and difficulty extracting features, leading suboptimal performance. study, we developed an improved sampling strategy that optimizes target distribution. proposed approach incorporates broad category information, where each sample assigned both a fine label converts distribution into This ensures model focuses on features corresponding category. We also introduce generation method same number categories in improve accuracy. experimental results demonstrate outperforms object methods. believe has potential be applied recognition, thus contributing development high-precision applications.