高光谱的菠菜新鲜度识别
本文以圆叶菠菜的高光谱图像作为研究对象。首先，获取了每片菠菜叶片在373-1034nm波长范围内的反射率，用自适应分组和自定义分组的精英选择策略结合菠菜高光谱图像反射率寻找能够较好的区分菠菜新鲜度等级的波长。利用支持向量机验证波长分类准确率，筛选出3个波长，分别为389.55nm,742.325nm,1025.662nm。从高光谱图像集从提取3个波长对应的图像样本，构成了4个样本图像库，分别为灰度图像库img389、img742、img1025和融合图像库img_merge,每个图像库含有681个图像样本。最后，基于深度学习技术建立菠菜新鲜度识别模型。图像样本库img389、img742、img1025和img_merge的识别准确率分别为85.94%、70.31%、67.19%、85.94%。通过对比发现基于389.55nm灰度图像和融合图像的菠菜新鲜度识别模型有更好的识别效果，模型最高测试集分类准确率达到85.94%。
目录
摘要 １
关键词 １
Abstract. １
引言
引言
1选题背景 ２
1.1 国内外研究状况 ２
1.1.1 国外研究状况 ２
1.1.2 国内研究状况 ２
1.2 技术介绍 ２
1.2.1 高光谱图像 ２
1.2.2 深度学习技术 ２
1.3 研究内容和技术路线 ４
1.3.1 研究内容 ５
1.3.2 技术路线 ５
2波长的选择 ６
2.1 数据预处理 ６
2.1.1 新鲜度等级划分 ６
2.1.2 数据去噪处理 ６
2.2 改良分组精英策略 ７
2.2.1 杂交与变异算子 ８
2.2.2 保护最优个体 ８
2.2.3 结果分析 ８
2.3 支持向量机验证 １０
3 建立图片数据库 １１
3.1 抽取3个波长下的灰度图像 １１
3.2 叶片图像分割 １２
3.3 图片归一化 １３
4 基于深度学习的菠菜新鲜 *好棒文|www.hbsrm.com +Q: ￥351916072￥
度检测 １４
4.1 搭建网络 １４
4.2 训练优化 １５
4.2.1 参数调整 １５
4.2.2 训练方式的调整 １７
4.3 测试结果分析 １７
4.4 可视化界面 １８
5 结论 ２０
5.1 总结 ２０
5.2 不足与展望 ２０
致谢 ２０
参考文献： ２１ 基于高光谱的菠菜新鲜度识别
Recognition of Spinach Freshness Based on Hyperspectral Images
Student majoring in network engineering HUANG Qiugui
Tutor XIE Zhonghong
Abstract： In this paper, hyperspectral images of spinach leaf were used as the research object. Firstly, the reflectance of each spinach leaf in the wavelength range of 3731034 nm was obtained. The elite selection strategy combining selfadaptive grouping and custom grouping was combined with the reflectivity of spinach hyperspectral images to find wavelengths that could better distinguish spinach freshness grades. . Using the support vector machine to verify the wavelength classification accuracy, three wavelengths were selected, namely 389.55 nm, 742.325 nm and 1025.662 nm. From the hyperspectral image set, the image samples corresponding to 3 wavelengths are extracted to form 4 sample image libraries, which are gray image library img389, img742, img1025 and fusion image database img_merge. Each image library contains 681 image samples. Finally, spinach freshness recognition model was established based on deep learning technology. The accuracy rates of image img389, img742, img1025 and img_merge are 85.94%, 70.31%, 67.19% and 85.94%, respectively. By comparison, the spinach freshness recognition model based on the 389.55 nm gray image and the fusion image has a better recognition effect. The classification accuracy of the highest test set of the model reaches 85.94%.
Key words: Hyperspectral; Elite selection strategy; Support Vector Machines; Deep learning
引言：我国是农业大国，种植业发展迅速。其中，蔬菜种植面积广，产量大，是我国的第一大农产品[1]。菠菜因其营养价值丰富，得到人们的广泛喜爱，销售范围较广。菠菜的新鲜度评级是其能否打开国内外市场的基本因素。菠菜采摘后，容易皱缩黄化，影响其品质和营养价值。所以菠菜采摘后若得不到及时的处理将导致损失增大，影响种植人员的收益。在实际生产中，人们往往依靠肉眼对菠菜的品质进行判断，效率低，判断误差大。因此,利用无损农产品检测技术进行无损、快速、准确地检测菠菜新鲜度是发展菠菜产业的关键技术。高光谱成像技术可以获取目标对象的光谱和空间信息，能够在不破坏菠菜叶片的情况下获得更多的信息[2]。因此，研究基于高光谱图像的菠菜新鲜度等级区分对促进菠菜的销售加工和提高菜农的收入具有重要意义。
目录
摘要 １
关键词 １
Abstract. １
引言
引言
1选题背景 ２
1.1 国内外研究状况 ２
1.1.1 国外研究状况 ２
1.1.2 国内研究状况 ２
1.2 技术介绍 ２
1.2.1 高光谱图像 ２
1.2.2 深度学习技术 ２
1.3 研究内容和技术路线 ４
1.3.1 研究内容 ５
1.3.2 技术路线 ５
2波长的选择 ６
2.1 数据预处理 ６
2.1.1 新鲜度等级划分 ６
2.1.2 数据去噪处理 ６
2.2 改良分组精英策略 ７
2.2.1 杂交与变异算子 ８
2.2.2 保护最优个体 ８
2.2.3 结果分析 ８
2.3 支持向量机验证 １０
3 建立图片数据库 １１
3.1 抽取3个波长下的灰度图像 １１
3.2 叶片图像分割 １２
3.3 图片归一化 １３
4 基于深度学习的菠菜新鲜 *好棒文|www.hbsrm.com +Q: ￥351916072￥
度检测 １４
4.1 搭建网络 １４
4.2 训练优化 １５
4.2.1 参数调整 １５
4.2.2 训练方式的调整 １７
4.3 测试结果分析 １７
4.4 可视化界面 １８
5 结论 ２０
5.1 总结 ２０
5.2 不足与展望 ２０
致谢 ２０
参考文献： ２１ 基于高光谱的菠菜新鲜度识别
Recognition of Spinach Freshness Based on Hyperspectral Images
Student majoring in network engineering HUANG Qiugui
Tutor XIE Zhonghong
Abstract： In this paper, hyperspectral images of spinach leaf were used as the research object. Firstly, the reflectance of each spinach leaf in the wavelength range of 3731034 nm was obtained. The elite selection strategy combining selfadaptive grouping and custom grouping was combined with the reflectivity of spinach hyperspectral images to find wavelengths that could better distinguish spinach freshness grades. . Using the support vector machine to verify the wavelength classification accuracy, three wavelengths were selected, namely 389.55 nm, 742.325 nm and 1025.662 nm. From the hyperspectral image set, the image samples corresponding to 3 wavelengths are extracted to form 4 sample image libraries, which are gray image library img389, img742, img1025 and fusion image database img_merge. Each image library contains 681 image samples. Finally, spinach freshness recognition model was established based on deep learning technology. The accuracy rates of image img389, img742, img1025 and img_merge are 85.94%, 70.31%, 67.19% and 85.94%, respectively. By comparison, the spinach freshness recognition model based on the 389.55 nm gray image and the fusion image has a better recognition effect. The classification accuracy of the highest test set of the model reaches 85.94%.
Key words: Hyperspectral; Elite selection strategy; Support Vector Machines; Deep learning
引言：我国是农业大国，种植业发展迅速。其中，蔬菜种植面积广，产量大，是我国的第一大农产品[1]。菠菜因其营养价值丰富，得到人们的广泛喜爱，销售范围较广。菠菜的新鲜度评级是其能否打开国内外市场的基本因素。菠菜采摘后，容易皱缩黄化，影响其品质和营养价值。所以菠菜采摘后若得不到及时的处理将导致损失增大，影响种植人员的收益。在实际生产中，人们往往依靠肉眼对菠菜的品质进行判断，效率低，判断误差大。因此,利用无损农产品检测技术进行无损、快速、准确地检测菠菜新鲜度是发展菠菜产业的关键技术。高光谱成像技术可以获取目标对象的光谱和空间信息，能够在不破坏菠菜叶片的情况下获得更多的信息[2]。因此，研究基于高光谱图像的菠菜新鲜度等级区分对促进菠菜的销售加工和提高菜农的收入具有重要意义。
版权保护: 本文由 hbsrm.com编辑，转载请保留链接: www.hbsrm.com/jsj/wljs/397.html
