Font Size: a A A

Research On Classification Of The Healing State Of Melon Grafted Seedlings Based On Hyperspectral Imaging

Posted on:2023-04-12Degree:MasterType:Thesis
Country:ChinaCandidate:J K YangFull Text:PDF
GTID:2532307118995709Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
The purpose of grafting is to improve the ability of plants to resist soil-borne diseases and abiotic stress.Melon is an important crop for grafting cultivation in production,and healing is an important part in the production process of grafted seedlings.In order to meet the needs of the industrialization development of seedling plants,it is urgent to develop a method for quickly judging the state of melon grafted seedlings during the healing period.However,traditional machine vision methods can only classify the healing state of melon grafted seedlings from the image level.When the traits are similar,the classification accuracy decreases.How to fully mine the data features that affect the classification of the healing state of melon grafted seedlings has become the focus of research at this stage.The main contents of this paper are as follows:(1)In order to fully extract the spectral features that are beneficial to the classification of the healing state of melon grafted seedlings,a global optimal strategy was designed by combining the local optimal strategies of different pretreatment and feature extraction.The strategy first performs spectral preprocessing on the original hyperspectral data,and selects the optimal spectral preprocessing algorithm by performing spectral analysis on different types of melon grafted seedlings with different days,and then performs spectral feature extraction on this basis.Before extraction,adding the difference feature information of grafting between living and non-surviving melon seedlings can further improve the classification accuracy on the basis of screening fewer bands.(2)In order to fully extract the spatial-spectral features that are beneficial to the classification of the healing state of melon grafted seedlings,by introducing the idea of deep learning and using the powerful feature extraction ability of convolutional neural networks,several groups of convolutional neural network models with different dimensions and different combinations are designed to carry out comparative analysis,that is,three-dimensional convolution kernel extracts spatial-spectral joint features,two-dimensional convolution kernel extracts spatial features,and one-dimensional convolution kernel extracts spectral features.The experimental results show that the three convolution kernel splicing models with different dimensions have the best classification effect,which verifies the idea of hyperspectral "spatial-spectral integration".(3)In order to further extract the spatial-spectral features that are beneficial to the classification of the healing state of melon grafted seedlings.On the basis of deep learning,the idea of attention mechanism is introduced,and a dual-branch multi-attention mechanism is designed based on the multi-dimensional convolutional neural network model.The model adds a channel attention mechanism and a spatial attention mechanism to the spectral branch and the spatial branch respectively,which can improve the weight of the output features of important channel features,and finally can extract deeper spectral features and spatial features.At the same time,due to the use of lightweight network,the model converges faster,effectively improves the calculation efficiency.In practical application,this study can conclude that the third day is the differential mutation days of grafted surviving seedlings and non-surviving seedlings.The state of grafted surviving seedlings can be divided into three stages:weak-medium-strong,and the state of non survival seedlings can be divided into two stages: weak-weaker.At the same time,it can be concluded that spectral characteristics are an important factor in the classification of melon grafted seedlings.Spatial characteristics do not contribute much to the classification of grafted surviving seedlings,but they make a great contribution to the classification of non-surviving seedlings.They can provide effective guidance for the production and early warning of melon grafted seedlings,and have a certain theoretical and practical value.
Keywords/Search Tags:Hyperspectral imaging, Healing state of melon grafted seedlings, Spatial-spectral integration, Convolutional neural network, Attention mechanism
PDF Full Text Request
Related items