Apple Multispectral Image Detection Based On Completely Local Binary Mode

China’s apple planting area is the largest country in the world,as well as the country with the highest output and consumption of apples.However,due to the lack of effective monitoring of quality changes in apple sorting,storage and transportation,the loss rate of apples in China after picking is as high as 20%,much higher than that in the United States and other developed countries.It is the premise of guaranteeing apple quality to realize high-efficient nondestructive testing in apple sorting,storage and transportation,but most of the domestic orchards still rely on manual sorting,supplemented by machine sorting,with high labor intensity and low sorting consistency.Facing the domestic to the requirement of modernization of all walks of life,the fruit industry has not only limited to comparing on production,but also want to change the mode of production,based on this,fruit nondestructive testing technology emerge in endlessly,development is rapid,more spectroscopy technique as a new technology has been successfully used in food,industrial and other fields,In this study,multi-spectral technology was combined with feature extraction algorithm and classification model to realize efficient recognition of skin lossy apple and skin lossy apple,and to explore the application of multi-spectral technology in fruit nondestructive testing.In this study,a multispectral data acquisition platform was built based on multispectral imaging technology,and the multispectral image information of apple in the range of 675～975nm was collected in 25 bands.Then the collected apple feature extraction,image information in local binary pattern and gradient histogram of these two kinds of traditional feature extraction algorithm based on the increase the local binary pattern of three kinds of improved algorithm including the complete local binary pattern,the partial derivative model and partial four values feature extraction algorithm for the image characteristics obtained by apple,based on the decision tree classification model,With average recognition accuracy,model running time and feature extraction dimension as evaluation indexes,the CLBP_S/M feature extraction operator in completely local binary pattern feature extraction algorithm was optimized to extract apple feature information.Will be extracted to apple characteristic information respectively through the K-nearest neighbor algorithm,random forest algorithm,support vector machine algorithm and the reverse structures,classification model,the neural network algorithm for different feature extraction algorithm based on CLBP_S/M operator under the skin damage to apple and spoil the multispectral image information recognition accuracy of the apple,the recall rate and specific degrees.Through the comparison of three evaluation indexes and the model running time,k-nearest neighbor classification algorithm is optimized as the optimal algorithm of this study.Under this classification model,the average recognition accuracy of 25 bands reaches 98.09%,and the model running time is 0.0769s.In this paper,multispectral images of apples are collected by multispectral technology,combined with complete local binary pattern feature extraction algorithm,which can effectively improve the recognition accuracy of classification model for skin damaged apples and skin lossless apples,and the classification time is short.Based on this,this study constructs a fast and accurate nondestructive apple recognition model,realizes the rapid recognition of damaged apples and non-destructive apples,and also provides some reference for the later application of multispectral technology in fruit nondestructive testing.Figure[38]Table[13]Reference[84]...