Automatic Grading And Taste Identification Of Fuji Apple Based On Image

China,as a big producer of apples,uses manual classification and mechanical classification in the link of apple sorting and quality testing at present.This way is time-consuming and laborious due to many human factors.Moreover,due to the inconsistent classification standards,the accuracy of the final detection results is not high.Therefore,how to conduct apple grading quickly,accurately and objectively has become the focus of research in the field of apple grading.In respect of apple quality,there is also carried out some research both at home and abroad,but mostly just stay in the inner components of the test and analysis for apple,such as hardness,PH,etc.And they have the high price of detection.For consumers,not only want appearance beautiful apples,but also want to get the high sweetness of the apple tasting.Therefore,how to judge whether apple is delicious by cheap,quick and nondestructive methods is also a problem to be solved urgently.In this paper,the most common Fuji apple type is selected for research.(1)An image acquisition device is built to obtain the full surface image of apples,which improves the problem of uneven brightness caused by Lambert phenomenon.At the same time,it improves the problem that the current apple grading study can rarely simulate manual detection.Then,the image size was calibrated to convert the pixel size to the actual size.In the part of image preprocessing,a new HSV color index is proposed to segment the foreground of apple image.(2)This paper presents a method to identify the defects and fruit stem/calyx,which improves the problem that the defects and fruit stem/calyx are easily confused in the top view and bottom view of apple.This paper designs algorithm for surface image of apple to better segment defect regions.This algorithm uses dual tree complex wavelet transform to extract the texture feature of target area,constructs support vector machine classifier to identify defects.And adding the decision tree method in the recognition algorithm is optimized,greatly shorten the recognition time,recognition accuracy of flaws and stem,flaws and calyx is 96.6% and 96.1% respectively.(3)A defect matching method based on Affine-SIFT(ASIFT)is proposed,which improves the problem that defects may be repeatedly detected during classification,removes the influence of repeated defects.This paper designs an algorithm to calculate the actual defect area.By building a fitting model,the error generated by mapping 3D objects to 2D images is greatly reduced.Finally,according to the requirements of "Red Fuji Classification",the decision tree idea was used to carry out classification,and the accuracy was 91%.The experiment showed that the classification method presented in this paper could achieve the classification of Fuji apples and meet the requirements of classification accuracy.(4)To improve the problem of weak texture of apple skin,an image enhancement method combining high-frequency emphasis filtering and histogram equalization is presented.The color feature and texture feature were fused and input into the multi-classification model of support vector machine for training.The parameter optimization was carried out by using particle swarm optimization algorithm,and the accuracy was 84.4%.The experimental results showed that the taste of Fuji apple could be identified by image processing technology.