Research On Apple Minor Damage Detection And Damage Time Estimation Technology Based On Hyperspectral Images

The output of China’s apple ranks first in the world,but the proportion of exports in total output is far lower than that of developed countries.The main reason is that the quality of finished apples in China is not up to standard and the waste of them is serious.Damage is a common factor affecting the quality of apples.When apples are damaged,they will decay quickly.With the extension of time,they will also affect other high-quality apples in the entire batch.The stability of the traditional artificial and chemical reagent detection methods of apple are poor and cannot meet the needs of actual production and life.Therefore,research on non-destructive testing technology for apple quality is of great significance for realizing rapid detection and classification of apple quality,reducing business losses,and improving market competitiveness.In view of the above problems,this paper has carried out the research on the detection of apple surface slight damage and the estimation of damage time based on hyperspectral images,and completed the following tasks:(1)In order to obtain complete and clear hyperspectral image data of intact and damaged apple,a hyperspectral image acquisition system was designed in this paper.Aiming at the problem that it was difficult to achieve uniformity when manufacturing apple damage,a set of slope type damage device was designed.(2)In order to solve the problem that the slight damage on the surface of the apple is not obvious and difficult to detect,this paper designed a method of locating the damage area based on the principal component transformation of the characteristic band.This method first used twice successive projections algorithm method to obtain the characteristic bands(821 nm and 940 nm)for detecting apple damage.Second,the principal components of the characteristic bands were analyzed.After analysis,the second principal component(PC2)with obvious difference from the damaged areas were selected as the effective image for detecting damage.Finally,the slightly damaged parts on surface of apple were segmented by threshold segmentation and edge detection methods.Experiments show that the overall detection accuracy of this method reaches 94.38%.(3)Aiming at the problem that using a single spectral feature to estimate the damage time is not ideal when the time interval of apple damage is short,this paper designed a method to estimation the apple damage time based on the spectral and texture features.This method first used MRMR-GA algorithm to extract spectral features(503 nm,554 nm,642 nm,679 nm,684 nm,715 nm,811 nm,864 nm,972 nm,978 nm,983 nm,989 nm)that were useful for estimating apple damage time,then the principal component analysis method was used to analyze feature images to obtain principal component images,and the principal component images were used to extract the texture features.Finally,the SVM,RF models were established and compared based on the combined features of the spectrum and texture.The results show that the estimation accuracy of the model based on combined features is higher than that of the model based on spectral features,and the RF model based on combined features has the best estimation effect,the accuracy is 93.93%.Through the analysis of the above experimental results,the principal component analysis method based on the feature bands designed in this paper can effectively detect the slight damage on the surface of apple.At the same time,the combination of texture features and spectral features can improve the accuracy of damage time estimation,make up for the lack of single spectral features.In conclusion,this study provides a theoretical basis and a technical reference for the development of apple-quality automated online testing equipment.