农产品品质性状信息获取机理与关键科学问题研究

农产品品质信息现场无损感知与实时质量检测是现代农业增产增收、品质评价、安全监控的重要手段。但由于农产品种类多、特性差异大和种植环境复杂，使得农产品品质性状信息现场无损感知较难。针对这一问题，本项目在前期农产品品质检测研究的基础上提出农产品品质性状信息获取机理与关键科学问题的研究，以浙江省特色水果--桃为研究对象，拟深入研究基于光学特性和嗅觉可视化的水果品质性状信息无损感知机理，研究感知器件与水果表面之间的互作机理，完成复杂环境下的水果品质性状信息的现场无损获取。最后，针对水果不同属性因子引起的预测模型普适性差的问题，结合深度学习和端到端特征学习模式，揭示不同水果属性因子对模型构建的影响机理，建立全局-局部模型耦合的学习机制。最终为建立浙江地区农产品现场无损感知新方法提供理论基础、方法和数据。

Quality information sensing and real-time quality monitoring of agro-products are important means of increasing production and income, quality evaluation and safety monitoring in modern agriculture. However, due to the variety of agro-products, large differences in characteristics and complex planting environment, it is difficult to perceive on-site information on the quality characteristics of agro-products non-destructively. In view of this problem, this project proposes the research on the mechanism of information acquisition and key scientific issues of agro-product quality traits on the basis of the previous research on their quality detection. Taking the characteristic fruit in Zhejiang Province - peach as the research object, it plans to study the mechanism of non-destructive sensing of fruit quality traits based on optical properties and olfactory visualization, study the interaction mechanism between the non-destructive perception device and the fruit surface, and complete the restoration fruit quality information acquisition in miscellaneous environment. Finally, aiming at the problem of poor universality of the model caused by different attribute factors of fruit, combined with deep learning and end-to-end feature learning mode, this project will reveal the influence mechanism of different attribute factors of agro-products on model construction, and establishes a global local model coupling multi task learning mechanism. Finally, it will provide theoretical basis, methods, and data for the establishment of a new method of non-destructive sensing of agro-products in Zhejiang Province.