Fruit Classification Based On Machine Vision Methodological Study

With the rapid development of the economy and the improvement of living standards,people have higher and higher requirements for the types and quality of fruits.The quality classification of fruits affects the price of fruits and people's consumption experience,while defects on the surface of fruits directly affect the quality grade and classification of fruits.At present,the quality inspection of fruits mainly depends on manual labor,which is time-consuming,labor-intensive,and unstable.In view of this,this paper designs a fruit detection system based on machine vision,researches and improves the fruit surface defect detection algorithm based on machine vision,which will help mechanized and automated operation of fruit surface defects non-destructive detection and quality classification to reduce Human workload and improve detection efficiency and accuracy.This thesis focuses on Crown Pear,Orange and Apple.The main research contents include:(1)Build a visual inspection system.Taking crown pears,oranges and apples as research objects,the design and experimental requirements of the fruit surface defect system were analyzed,and the selection of cameras,lenses,light sources,light boxes,and computers in the machine vision inspection system was completed,and a fruit surface quality inspection system was established.(2)Fruit surface image preprocessing.Preprocess the fruit surface image according to the experimental needs,and use the gray method of component method,maximum method,weighted average method,and average method to generate grayscale images.Select the weighted average method with more obvious difference between samples and background as the gray.Degree processing method.The piecewise linear function is used as the image enhancement method to make the image defect area more prominent.The two methods of 2D OTSU threshold segmentation and the improved 3D OTSU threshold segmentation are compared.The results show that the 3D OTSU threshold segmentation is more accurate for defect boundary segmentation.After comparing several commonly used edge detection operators,a canny operator with good anti-noise interference and accurate positioning is selected as the edge detection method.(3)Fruit surface feature value extraction.By analyzing the geometric and color characteristics of fruit,three common fruit defect characteristics were identified,which improved the accuracy of fruit surface defect recognition and quality classification.Starting from the texture features,the gray surface co-occurrence matrix and the improved SURF algorithm were used to analyze the characteristics of fruit surface defects.The energy,entropy,contrast and contrast moment of the four characteristic parameters of the gray level co-occurrence matrix are calculated.The results show that energy can be used as an effective eigenvalue of defect classification.At the same time,in the feature point matching stage,the nearest-neighbor vector matching algorithm and random sampling algorithm are used to improve the SURF algorithm.An improved SURF algorithm was used to analyze and calculate the surface texture features of the fruit.Experiments were performed on the gray level co-occurrence matrix and the improved SURF algorithm.The results show that the improved SURF algorithm improves the speed of feature point extraction and matching accuracy,significantly reduces the matching time,and the accuracy of fruit surface defect recognition is better than the gray level co-occurrence matrix.(4)Classification of fruit surface defects.SVM classifier and improved convolution neural network classifier were used to study fruit surface defect recognition.The classification accuracy of SVM classifier for crown pear,orange and apple is 92.48%,92.95%,92.2%.The improved convolution neural network was used to construct three types of fruit automatic classification network models.The supervised learning method was used to perform sample training.The residual surface network was used to detect and classify the fruit surface feature values.The classification accuracy was 95% and 95.625,respectively 96.25%.The analysis of the experimental results shows that the improved convolution neural network algorithm proposed in this study is feasible,the classification efficiency of fruit surface defects is high,the robustness is good,and the classification accuracy is significantly higher than the SVM classifier.