| Snow pears inevitably suffer diseases and injuries during their growth,picking,transportation,storage,and sales.Testing and classifying damaged snow pears can screen for pathogens and isolate them,thereby reducing the spread of the disease and reducing the food safety risks associated with it.In addition,setting corresponding prices and uses based on the degree of damage to snow pears can generate significant economic benefits.Multi spectral Image(MSI)technology has the advantages of efficient,non-destructive,accurate,and reusable detection.This study is based on MSI combined with artificial intelligence algorithms to accurately detect and classify snow pear injury,and further subdivide the degree of damage.The multi-spectral images of intact,diseased,and damaged snow pears are the research object of this study.The damaged samples are manufactured using a quantitative damage device,and the average spectral values of the area of interest of the damaged snow pears are selected as the dataset for injury classification.The dataset is preprocessed,anomaly detected,and sample set divided.Then support vector machine(SVM),random forest(RF),adaptive boosting(Ada Boost),convolutional neural network(CNN)with two-layer convolution structure The 18 layer Residual Neural Network(Res Net18)and34 layer Residual Neural Network(Res Net34)models were used to classify injuries and diseases in Snow Pear.After detecting and classifying injuries,remove the sample data of the diseased snow pear,and then calculate the Euclidean Distance(ED)between the average spectral values of snow pear with different degrees of damage.Based on the relative position of ED in the spatial coordinate system and the clustering analysis results of K-Means Clustering Algorithm(K-Means),the damage degree of snow pear is comprehensively subdivided.Evaluate the performance of the above model using a test set,and evaluate the model based on its classification accuracy and Loss value.The classification results show that the best model for detecting and classifying snow pear injury is Res Net34,with a test set accuracy of 0.9706 and a Loss value of 0.0952;The Ada Boost model performs well in classifying the degree of damage to Snow Pear,with a detection accuracy of 0.9316 and an F1 value of 0.9217.On this basis,this study developed an intelligent detection and classification system for snow pear injury,which can upload the collected sample data in real-time to the established model and effectively provide prediction and classification results for snow pear injury.The research results indicate that MSI can accurately,effectively,and quickly identify damaged snow pears,and subdivide snow pears with different degrees of damage.This provides a new approach for the practical application of MSI based intelligent detection and classification of snow pear injuries. |
