Research On Kiwifruit Quality Grading Based On Image Multi-feature Fusion And Neural Network

Kiwifruit is a fruit with rich nutritional value and medicinal value,but it is susceptible to natural decay during storage,resulting in a decline in quality,affecting consumer satisfaction and market value.Effective postnatal treatment of Kiwifruit can minimize these losses and ensure the quality and taste of Kiwifruit.Postnatal processing of Kiwifruit includes a series of processes such as cleaning,grading,packaging,and storage.Among them,kiwifruit grading is a key step after fruit picking and before entering the market for sale,which is used to ensure fruit quality,improve its added value,and increase market competitiveness.Therefore,the grading treatment of kiwifruit plays an important role in improving the market competitiveness of Kiwifruit,meeting consumer demand,and increasing the added value of the kiwi industry.Currently,kiwifruit is traditionally graded in China by hand and machine,both of which have certain limitations: on the one hand,the results of manual grading are prone to subjectivity,making it less accurate and high cost;On the other hand,machine grading can easily cause damage to fruits.Automatic grading of Kiwifruit based on computer vision technology can effectively avoid the subjectivity of manual grading and fruit damage caused by machine grading,which has important practical value and economic significance.In this paper,we explore automatic grading detection methods for kiwifruit based on image multi-feature fusion and neural networks,taking the fruit of the "Cui Xiang" kiwifruit as the object of study and combining current research results in this field.Considering the importance of image background for image data recognition and classification,this article uses black as the background color when photographing Kiwifruit.Firstly,when performing preprocessing tasks such as graying,filtering,and image cutting on Kiwifruit images,based on analyzing traditional image cutting methods,the S component of HSV(Hue,Saturation,Value)color space and grayscale information are combined to separate the Kiwifruit part from the background.Secondly,in terms of extracting appearance features such as color,texture,shape,size,and defects of kiwifruit,the S component of the HSV color model and the color moments of RGB(Red,Green,Blue)colour model to represent its colour features;the epidermal texture was described using the equivalence model LBP histogram;for the shape size of the kiwi,calculation of specific data such as area,circumference,long axis,short axis and aspect ratio on the basis of traditional studies;the defective regions were segmented using a threshold segmentation method,and the area and colour moments of the defective regions were used to describe the kiwi defective features.Subsequently,single feature and fused feature grading experiments were designed based on the extracted features,and the fused feature model was shown to have stronger grading performance.Finally,to address the limitations of machine learning methods for grading,deep learning techniques were used to grade kiwifruit quality.Based on the study of Res Net and Vision Transformer model,the two were fused and improved to build the Res TNet model,and the model was indicated to be effective in improving the accuracy of kiwifruit grading through experiments.