Detection Technology Of Fruit Quality And Fruit Tree Leaf Health Based On Hyperspectral Imaging

Fruit and foliage are two important components of a fruit tree,and the health of the foliage and the quality of the fruit reflect the growth and development of the tree.The leaves of fruit trees are susceptible to viruses and bacteria due to various factors such as climate and management,which can affect the development of the fruit.Fruit is an economically important agricultural product and non-destructive testing of fruit quality and foliage is important for the development of agriculture.Hyperspectral imaging is an emerging non-destructive testing(NDT)technology that is widely used in agricultural monitoring and quality control of agricultural products.Research into hyperspectral NDT techniques has focused on extracting valid key information from complex data,with most suffering from a lack of consideration of data correlation,insufficient accuracy and inadequate use of data.The existing data extraction methods can be divided into waveband selection methods and spatial feature extraction methods according to the different data objects under study.In the band selection method,correlation analysis between bands is added to the existing band selection method,and the application scenarios of the algorithm are broadened.In the spatial feature extraction method,the effective fusion of spatial and spectral information is investigated and combined with deep learning to be applied to classification tasks.The main contributions of this paper are as follows,using Korla fragment pears and the leaves of loquat trees as research objects:(1)To address the problems that the MIA band selection algorithm lacks band correlation consideration and cannot be adapted to quantitative analysis,the Gs MIA band selection method was proposed and applied to the study of quantitative analysis of the sugar content of Korla fragment pear.The correlation between different bands in the spectral data was analysed,and the Gs MIA band selection algorithm was proposed to select the key bands in three stages.Firstly,the most discriminative bands were selected using the Relief F algorithm,and the selected bands were used as clustering centres to group the full bands using the k-means algorithm;then the importance of each band was calculated using the modified MIA algorithm in groups to initially select the key bands;finally,the redundant bands were removed using the UVE algorithm to complete the selection of the key bands.The SVR prediction model was established using the key bands to achieve quantitative prediction of the sugar content of Korla fragment pear.(2)To address the problem that it is difficult to improve the classification accuracy of traditional classifiers,a method based on waveband selection algorithm and artificial neural network is proposed to diagnose the health condition of fruit tree leaves.Taking loquat leaves as an example,the collected loquat leaves are classified into four categories according to their maturity and health status.These are: three artificial neural networks based on the full band,a machine learning classification method based on the feature band and an artificial neural network classification method based on the feature band.The experimental results demonstrate that the combination of the band selection method and the artificial neural network gives better results and saves computational resources.(3)A fruit tree leaf diagnosis method based on spatial-spectral fusion technology and transfer learning to address the problem of inadequate use of hyperspectral image data.The leaves of loquat trees were classified into three levels according to the degree of disease.The first three principal components of the image were extracted using PCA feature extraction and assigned to RGB channels to synthesise a false colour image.To solve the problem of inadequate data sets and varying image sizes,the data set was expanded by cutting the images to equal size and data enhancement was performed.A VGG-16 pre-trained model was used and fine-tuned to achieve a hierarchical diagnosis of loquat leaf diseases using transfer learning techniques.