Research On Classification And Grading Technique Of Apple Based On Multi-features

China encounters some predicaments such as great apple quality varying,low price in markets,although it is the world largest apple producer.And these lead to relatively low economic benefits across the global market.Reasons of this situation may be shortage of post-harvest fruit processing technology and poor ability of automated grading and filtering.Therefore,there is inevitable and heavy demand for apple industry to improve post-harvest processing technology to achieve optimization grading for apple quality.With the rapid development of computer vision,image processing technology matures for achieving apple appearance-based quality grading,which has advantages such as high speed and precision,nondestructive detection,reducing labor force and errors,unification and standardization,and which can gradually replace the traditional apple grading methods,e.g.manual grading and mechanical grading.In this paper,we investigated the research work including single feature,multi-features on apple grading technology and other grading of agricultural products during the past two decades: from 1990's to 2015.This paper presents a new method for apple multi-features grading: Firstly,we apply homography image calibration and pre-processing for sample images of 80 apples(4 images per apple).In the HIS color space,images can be segmented using OTSU method by comparing histograms of H,S and I components and Canny edge detection is used to extract the edge of the apple.Then,maximum diameter of the cross section of the apple profile image is extracted as size feature,circularity as the shape feature,and the proportion of red region on apple surface is calculate as color feature.Also,we extract defect area and get the calculated area size as defect feature value.Next comes grading apples which means dividing apples with defection into the second class and third class according to defect feature value and national grading criteria.For apples without defection,three kinds of weights of 0.35,0.18,0.47 were set for the size,shape and color feature by using entropy method combined with experimental results.Finally,K-means clustering algorithm is applied to classify apples into four classes obtaining the weights of those three features and four cluster centers are calculated.By comparing the Euclidean distance between values of the apple samples to be graded with the four cluster centers,the multi-features grading of apples can be realized.Results show that the total accuracy of the proposed multi-features grading method of 80 apple samples is 95%,and the processing speed is 7 images per second.Compared with other grading methods,this method is more efficient,convenient and precise,which is enough to meet the current industrial requirement of apple grading.Research of this paper provides technical support and research direction for multi-feature automatic grading of fruit industry.