Detection Of Slight Damage Of Akesu Rock Candy Apple Based On Hyperspectral Image Technology

Apple surface damage detection is a very important link in the development of Akesu rock candy apple industry.Apples are prone to collision damage during picking and transportation.Timely detection and elimination of the collision damage can effectively prevent the expansion of the damage area and avoid infecting normal apples.Therefore,it is particularly important to identify and detect the early slight damage of apples.In this paper,Aksu rock candy apple as the research object,Aksu rock candy apple belongs to Fuji class,is one of the most common apple species in China,the surface color has red and yellow,slightly damaged when the naked eye is difficult to detect,has a certain representativeness.In this experiment,300 samples from three batches of bingtangxin apples were purchased for treatment,including 90 damaged samples,60 stem / calyx samples and 150 normal samples for reference.Aiming at the situation that the slight damage is difficult to identify and the degree of damage can not be determined,based on hyperspectral image technology,this paper uses the spectrum fusion processing method to identify and detect the surface collision damage of apple(1)254 bands of Apple spectral data were collected by Hyperspectral camera.Principal component analysis(PCA)was used to reduce the dimension of full band spectral image.The principal component(PC)image with obvious difference between normal area and damaged area was extracted by using the covariance matrix and correlation coefficient matrix algorithm of PCA The method of threshold segmentation and gray level morphological transformation is used to identify the damage area and extract it.After many experiments,the accuracy of damage identification is more than 96%.(2)The hyperspectral images of damaged area,normal area and peduncle / calyx area were collected again.The preprocessed hyperspectral images were stratified by gray level,and then the contour lines of equal gray level were analyzed.The peduncle / calyx area was identified by RGB pseudo color enhancement.It is proved by many experiments that the recognition accuracy of fruit stem / calyx region is more than 97%.(3)An estimation algorithm based on hyperspectral imaging technology is used to evaluate normal apples,slightly damaged apples and severely damaged apples.The algorithm involves: resampling the damaged area to obtain as much damage information as possible;In principal component regression analysis,two principal components which can contain most or main information are selected for regression modeling;The partial least squares(PLS)regression model was established.After cross validation,the interpretability R2 value of the treatment model results was obtained,which can be used to evaluate the damage degree.The research results show that the fusion processing method used in this paper can effectively identify the slight damage of apple,and establish a model to evaluate it,which can provide an effective scheme for the automatic production line,provide a theoretical basis for apple quality detection,and contribute to the improvement of fruit automatic classification and export rate.