Rapid Detection Of Canola Rapeseed Seedlings Based On Computer Vision Research

In recent years,the dual-purpose utilization of rapeseed,known as "oil and vegetable," has achieved significant development and application in China.Rapeseed sprouts,a nutritious and popular green vegetable,have gained consumer appreciation.However,after harvest,rapeseed sprouts undergo a series of physiological and chemical changes,leading to a decline in their quality and loss of both nutritional and commercial value.In order to ensure the food safety of vegetables for consumers and enhance the market competitiveness of rapeseed sprouts,there is an urgent need to develop a rapid and non-destructive detection method for assessing the freshness of harvested rapeseed sprouts.Therefore,this study focuses on the dual-purpose rapeseed variety "Shishan cai tai" and employs computer vision combined with machine learning methods to investigate the rapid detection of freshness in rapeseed sprouts stored with and without bagging.The main contents of this study are as follows:(1)By measuring the moisture content,hardness,chlorophyll a,chlorophyll b,and vitamin C of rapeseed sprouts during storage and combining sensory evaluation,the quality changes of rapeseed sprouts under different storage conditions are studied.Through comprehensive analysis of six freshness evaluation indicators,a freshness grade system for rapeseed sprouts under different storage conditions is established,providing a research foundation for subsequent freshness detection.The research results indicate that using cluster analysis and principal component analysis,the freshness of rapeseed sprouts can be divided into three grades.For the storage condition without bagging,the first day is considered fresh,the second to third days are considered slightly fresh,and the fourth to fifth days are considered not fresh.For the storage condition with bagging,the first to third days are considered fresh,the fourth to sixth days are considered slightly fresh,and the seventh to eighth days are considered not fresh.(2)An image acquisition device is constructed to capture leaf surface images of rapeseed sprouts under different storage conditions.The images are preprocessed using bilateral filtering combined with the marker-based watershed algorithm,and texture features of the segmented images are extracted.Based on the dimension-reduced feature values obtained through the GA-ANN algorithm,SVM,K-NN,and ANN models are separately established to assess the freshness grades of rapeseed sprouts.The performance indicators of the three models are compared,and the model with the best classification performance for freshness detection of rapeseed sprouts is selected.The research results show that for the storage condition without bagging,the ANN model exhibits the best classification performance with accuracy,precision,recall,and F1-score reaching 91%,91.01%,92.5%,and 91.64% respectively.The analysis of the confusion matrix indicates that the model can effectively discriminate the freshness grades of rapeseed sprouts under the storage condition without bagging.For the storage condition with bagging,compared to the SVM and K-NN models,although the ANN model also achieves the best classification performance,its accuracy,precision,recall,and F1-score are only 72.5%,74.37%,71.67%,and 72.55% respectively,indicating a moderate classification performance for the freshness grades of rapeseed sprouts under the storage condition with bagging.(3)For the rapeseed sprouts stored with bagging and considering the lower classification accuracy of traditional machine learning models,the study explores the application of deep learning methods based on convolutional neural networks.Specifically,VGG16 and Res Net50 models are separately established to assess the freshness grades of rapeseed sprouts under the storage condition with bagging.To further improve the classification performance,the Res Net50 model is enhanced with the SE attention mechanism.The research results demonstrate that the performance of the Res Net50 model is superior to that of the VGG16 model,with an accuracy of 94.17%,precision of 94.64%,recall of 94.35%,and F1-score of 94.15%.The improved Res Net50 model further enhances its performance,achieving an accuracy improvement of 1.87%,precision improvement of2%,recall improvement of 1.94%,and F1-score improvement of 1.83% compared to the original model.The improved model effectively discriminates the freshness grades of rapeseed sprouts under the storage condition with bagging.