本项目研究基于迁移学习理论的癌症高光谱影像分类的关键理论和方法，面向该领域中存在样本负迁移、类不平衡、数据和模型异构、模型缺乏可调试性等问题展开研究。研究内容包括癌症高光谱影像迁移学习中的样本不确定性问题研究；癌症高光谱影像迁移学习中的数据源和模型异构问题研究；癌症高光谱影像迁移学习模型的可解释性问题研究。针对这些研究内容，本项目提出了使用深度强化学习、采样模型融合代价函数、虚拟样本生成、Drop-boosting模型参数迁移、样本影响函数和特征空间分布影响函数的方法。在本课题理论研究的基础上，研究成果将应用于癌症高光谱影像智能分析的领域，作为医生重要的辅助手段提升癌症诊断水平。..本课题旨在突破现有基于深度学习方法需要大量规整训练数据的局限性，探索在人工智能和医疗影像智能分析这个交叉领域的新理论和新方法，为基于人工智能技术解决医疗工作中的重大需求提供新理论方法和可靠技术手段。

This project studies the key theories and methods of medical hyperspectral image classification based on Transfer Learning Theory, which faces the problems of few number of samples, negative transfer, unbalanced data, heterogeneity of data and model, lack of debug nature of the model, etc. The research includes the research of sample uncertainty in medical hyperspectral image transfer learning, the data source and model heterogeneity in medical hyperspectral image transfer learning, the interpretability of medical hyperspectral image transfer learning mode . In view of these research contents, this project proposed methods of using depth-enhanced learning, sampling model fusion cost function, virtual sample generation, Drop-boosting model parameter transfer, sample influence function and feature space distribution impact function. On the basis of these theoretical studies, the results of this project will be applied to the field of intelligent analysis of cancer hyperspectral images as an important aid to enhance the level of cancer diagnosis..The purpose of this project is to break through the limitations of the existing training data that require a large amount of standardization based on the deep learning method and to explore new theories and new methods in the field of artificial intelligence and intelligent analysis of medical images in order to solve the problem that artificial intelligence technology is used in medical work and provide new theories and robust techniques.