Research On Nondestructive Detection Method Of Apple Mechanical Damage Based On Hyperspectral Imaging Technology

When picking,handling and storing apples,they are vulnerable to impact,crushing and vibration,which can cause mechanical damage.Mechanical damaged apple fruit taste and quality is reduced,susceptible to microbial infection,resulting in fruit rot and shortened shelf life,which seriously reduces the economic benefits of the apple industry.Nondestructive testing of apple mechanical damage and identification of the type and time of mechanical damage can not only timely eliminate severely damaged fruit,but also guide each production link to take scientific and effective protection measures,to a certain extent reduce the loss of post harvest fruit and maintain fruit quality,promote the healthy development of the apple industry.The mechanical damage factors mainly include impact damage,extrusion damage and vibration damage.The damage mechanisms of the three factors are different,which lead to differences in map feature detection.Based on the above analysis,this paper takes Apple as the research object and uses hyperspectral imaging as the technical means to carry out the nondestructive detection of mechanical damage on apple,including the detection of mechanical damage area,the identification of different types of mechanical damage and damage time.The main research contents of this paper are as follows:(1)The detection method of mechanical damage area of apple based on hyperspectral imaging technology is studied.After the collision,extrusion and vibration damage samples were artificially prepared,the image data of intact and damaged apples were respectively collected by hyperspectral imaging technology,and spectral and image analysis methods were used to achieve segmentation detection of damaged areas.Piecewise principal component analysis(PCA)and minimum noise separation transform(MNF)were used to reduce the dimensionality of the data.The optimal spectral range for damage detection was determined in the near infrared band range.Median filtering,adaptive threshold segmentation,and morphological operations such as expansion and corrosion were carried out on PC2 and MN2 to segment the damage region effectively.The results show that the average accuracy of PCA method is 92.1%,and the correct recognition rates of no damage,collision damage,squeeze damage and vibration damage are 98.3%,91.6%,83.3% and 95.0%,respectively.On the whole,PCA has a higher recognition rate for samples than MNF,while vibration damage has a higher recognition rate than compression damage and collision damage.Both PCA and MNF-based damage region detection methods can effectively detect three damage regions of different damage types.(2)The recognition method of different mechanical damage types of apple based on hyperspectral imaging technology is studied.Based on the characteristics of hyperspectral data,different types of mechanical damage were identified from the perspective of image and spectrum.From the perspective of image,PC2 images with obvious damage area were obtained,and pixel and texture features were extracted by directional gradient histogram and gray co-occurrence matrix.A mechanical damage type recognition model based on image features was established by support vector machine(SVM)and probabilistic neural network(PNN).From a spectral perspective,the spectra of the damaged areas detected by the Apple mechanical damage region detection method are extracted.Based on the feature wavelengths extracted by the non-information variable elimination method and the stability competitive adaptive Reweighted sampling method(SCARS),SVM and PNN mechanical damage type recognition models based on spectral features are established.The recognition effect of the model based on image and spectral features is compared.The results show that the recognition rate of the prediction set based on the image features extracted from PC2 is lower than 78.3%,while the recognition rate of the model based on the spectral features is higher than 84.1%.The overall effect of the recognition model based on the spectral features is better than that of the image features.Among the damage type recognition models based on spectral features,the SVM model based on SCARS extracted feature wavelengths has the best recognition effect,with overall recognition rates of 96.1% and 91.5% for the correction set and prediction set,respectively.The recognition rates of non-damage,extrusion damage and vibration damage are all 100.0% in the established optimal model,while the recognition effect of collision damage is poor.(3)The recognition method of apple damage time based on hyperspectral imaging technology is studied.Spectral features show excellent performance in the recognition of mechanical damage types,so the damage time recognition is only based on spectral features.The spectral data of apples damaged by collision,extrusion and vibration were collected at 3h,12 h,24h and 36 h after damage.The characteristic wavelengths of single damage and mixed damage samples were extracted,and the damage time recognition models based on single damage and mixed damage were constructed respectively.The results show that for the collision and extrusion damage samples,the recognition accuracy of all the established models for the four time periods is high,the average recognition rate of the correction set and the prediction set is higher than 86.4% and 81.2%,respectively.For the vibration damage samples,the model recognition results are significantly lower than the other two damage types.Compared with the damage time identification results of different sample sets,the recognition rate of mixed damage time identification model is slightly lower than that of collision or extrusion damage time identification model,but the model is more adaptable.The average recognition rate of collision and crush damage correction set and prediction set is 100.0%,and the average recognition rate of mixed damage prediction set is 94.4%.It indicates that the hyperspectral imaging technology can not only accurately identify the different damage time of collision and extrusion damaged apples,but also identify the different damage time of mixed damaged apples.