葡萄糖含量预测与多光谱对齐和改进的残余网络预测.
Yiming Chen1, Jizhou Deng1, Zhijie Liu1
1Hunan Agricultural University, Changsha, 410000, China.
Scientific reports
|October 22, 2025
概括
这项研究开发了一个Improved-Res深度学习模型,用于使用多光谱成像进行非破坏性葡萄糖含量检测. 该模型显著提高了预测准确性,超过了葡萄质量评估的传统方法.
科学领域:
- 农业工程 农业工程
- 计算机视觉 计算机视觉
- 频谱学是一种光谱学.
背景情况:
- 葡萄糖含量对于成熟和分级至关重要.
- 无接触,无破坏性检测对于自动化系统至关重要.
- 谱学为此提供了一个关键技术.
研究的目的:
- 开发一种非破坏性的方法来检测阳光葡萄中的糖含量.
- 预处理多光谱图像以检测噪声和错位.
- 构建和评估一个深度学习模型,以准确预测糖含量.
主要方法:
- 收集了2,880张葡萄的多光谱图像.
- 应用高斯反和增强相关系数 (ECC) 注册用于预处理.
- 开发了一个基于ResNet-50的改进分辨率模型,其中包含SE关注,DSC和Inception模块.
主要成果:
- 改进分辨率模型实现了0.49的平均平方误差 (MSE),0.55的平均绝对误差 (MAE) 和0.92.92的R平方 (R2).
- 这显著超过了传统的机器学习 (XGBoost:MSE=1.35,MAE=0.90,R2=0.78) 和标准的ResNet-50 (MSE=0.95,MAE=0.96,R2=0.84).
- 废弃实验验证了SE注意力,深度可分离卷积和Inception模块的有效性.
结论:
- 拟议的改进分辨率模型为非破坏性葡萄糖含量检测提供了一个高度准确和强大的解决方案.
- 这项技术可以提高葡萄采摘机器人和分拣平台的效率.
- 集成先进的深度学习技术对精准农业有很大的前景.
相关概念视频
01:10Sugars as Energy Storage Molecules
Sugar (a simple carbohydrate) metabolism (chemical reactions) is a classic example of the many cellular processes that use and produce energy. Living things consume sugar as a major energy source because sugar molecules have considerable energy stored within their bonds. Consumed carbohydrates have their origins in photosynthesizing organisms like plants. During photosynthesis, plants use the energy of sunlight to convert carbon dioxide gas into sugar molecules, like glucose. Because this...
01:10Sugars as Energy Storage Molecules
Sugar (a simple carbohydrate) metabolism (chemical reactions) is a classic example of the many cellular processes that use and produce energy. Living things consume sugar as a major energy source because sugar molecules have considerable energy stored within their bonds. Consumed carbohydrates have their origins in photosynthesizing organisms like plants. During photosynthesis, plants use the energy of sunlight to convert carbon dioxide gas into sugar molecules, like glucose. Because this...
01:21Precipitation Titration Curve: Analysis
The precipitation titration curve demonstrates the change in concentration of one reactant with the volume of titrant added. During the titration of chloride ions with silver nitrate, the precipitation titration curve is divided into three regions: before, at, and after the equivalence point. Before the equivalence point, low redissolution of the sparingly soluble silver chloride precipitate gives a low silver ion concentration. However, in the second region, representing the equivalence point,...