Research On Nondestructive Detection Of Apple's Internal Defects

Although china is a big country in apple production and consumption,in terms of export trade,the export rate is far below the world average.High-quality apples are in short supply.Many apples with internal defects such as moldy core,bitter pit,bruises and other internal defects are mixed in the apple market,resulting in a low rate of excellent fruit and causing great losses to fruit farmers.Because the internal diseases of apples cannot be identified from the appearance,this study proposes a non-destructive testing method for the detection of internal diseases of apples.The main content and conclusions are as follows:(1)According to the internal defects of the apple,the apple CT image is preprocessed to complete the operation of the apple sub-image acquisition,apple core extraction and injury area location.First,segment the foreground and background of the CT image based on the gray histogram.Using morphological denoising,maximum between-class variance method and connected domain labeling methods to complete the acquisition of apple sub-images.After that,the apple radius is obtained through the edge extraction of the apple.Apple's centroid coordinates and radius are used to extract the core of the apple.Finally,algorithms such as the maximum between-class variance method are used to locate the injured area.(2)According to the location of the apple core area,the moldy core disease identification is realized.The process of moldly core apples recognition is to use the maximum between-class variance method to threshold the fruit core image;Position the apple core area and apple seed area through the smallest enclosing rectangle;By calculating the percentage of the apple seed area to the core area,judge whether the percentage of the apple seed area to the core area is higher than 50%,if it is higher than 50%,the core is a moldly core,otherwise it is a healthy core.(3)Aiming at the location and recognition of injury areas,a recognition algorithm for bitter pit and bruise is proposed.First,perform feature extraction of the injured area,extract its shape feature,texture feature,and location feature,a total of 18 types of feature information;after that,the effective feature information is selected,and the two methods of multiple stepwise regression and the class distance separability criterion are used to perform effective feature selection on 18 types of feature information in the injured area.Common feature information is used as effective feature information;finally,perform the recognition of bitter pit and bruise.According to the selected effective feature information,use the support vector machine with default parameters and the support vector machine optimized by genetic algorithm to recognize bitter pit and bruise.The results show that the support vector machine optimized by the genetic algorithm has an accuracy rate of 95%for the recognition of bitter pit and bruise,which is significantly better than the support vector machine with default parameters.