目录
摘要II
关键词II
AbstractIII
引言
引言1
1 绪论1
1.1 植物叶片表型研究1
1.1.1 植物表型1
1.1.2 植物叶片表型1
1.2 植物叶片识别研究意义1
1.3 植物叶片识别研究现状2
1.4 课题内容2
1.5 章节安排2
2 数据集与预处理3
2.1 数据集3
2.2 预处理方式3
3 植物叶片识别算法与结果分析3
3.1 相关算法原理分析3
3.1.1 植物叶片切割 3
3.1.2 moblilenetV3神经网络 5
3.1.3 VGG 7
3.1.4 Alexnet 8
3.1.5 多任务学习 9
3.1.6 科目优先策略 9
3.2 植物叶片识别算法 10
3.2.1 植物叶片分割 10
3.2.2 网络训练模块 11
3.3 结果分析 12
4 植物叶片识别系统13
5 总结与展望14
致谢15
参考文献15
基于科目优先策略的植物叶片识别技术研究与应用
摘 要
植物叶片识别是识别植物种类较为方便的方式之一，其数据来源是植物数据中较为容易取得的，识别方式较为直观且符和大众辨别植物种类的行为习惯。实现对植物叶片的识别，对植物图像进行预处理，对预处理后的图像进行特征抽取，对植物的种类进行识别，最后与其他算法进行比较。程序采用grabcut对植物图片进行交互式切割，在输入复杂背景时仍然能够通过用户自行标注目标所在的矩形区域对图像进行分割，保证了图片输入深度学习网络后的识别准确率。模型采用科目优先策略，引入了多任务学习方式，通过对植物叶片的科目标签和种类标签进行识别与训练，降低网络过拟合的可能性，提高了网络对目标特征的识别能力，使得网络能够在数据量比较低的条件下，仍然具有较好的 *51今日免费论文网|www.51jrft.com +Q: ^351916072# 
训练结果。网络结构采用mobilenet模型，该模型为轻度模型，能够更好的移植到其他移动端平台，且相较于其他CNN网络运算所需的资源较少。
RESEARCH AND APPLICATION OF PLANT LEAF RECOGNITION TECHNOLOGY BASED ON FAMILY PRIORITY STRATEGY
ABSTRACT
Plant leaf identification is one of the more convenient ways to identify plant species. Its dataset is more easily obtained from plant data. The identification method is more intuitive and conforms to the publics habits of identifying plant species. The program aims to realize the identification of plant leaves, preprocess the plant image, extract the feature of the preprocessed image, identify the type of plant, and finally Compare with other algorithms. The program uses grabcut to interactively cut plant pictures. When entering a image with complex background, the image can still be segmented by the rectangular area where the user marks the target, which ensures the accuracy of recognition after the picture is input to the deep learning network. The model adopts a subjectfirst strategy and introduces a multitask learning method. By identifying and training the subject labels and category labels of plant leaves, the possibility of overfitting the network is reduced, and the networks ability to recognize target features is improved, making the network When the amount of data is low, you can still get better training results. The network structure adopts the mobilenet model, which is a mild model, which can be better ported to other mobile terminal platforms and requires fewer resources than other CNN network operations.
KEY WORDS：deep learning；convolutional neural；plant image recognition；image segmentation

原文链接：http://www.jxszl.com/jsj/wljs/606907.html