Research On Maturity And Damage Identification Of Fresh Pepper Based On Hyperspectral Technolog

China’s pepper production ranks first in the world,the traditional manual-based pepper harvesting method,time-consuming and laborious,can not meet the needs of industrial development,it is urgent to realize the mechanization,industrialization and intelligence of the pepper industry.The key technology of intelligent pepper harvesting is the identification of fresh peppers in the field.On the other hand,pepper will be squeezed and collided in picking,transportation and storage,once the damage is done,if not found immediately,the same batch of pepper will be infected,and then accelerate the decay of the whole batch of pepper.In response to the above problems in the pepper industry,this paper proposes a method for identifying the maturity of fresh peppers in the field and their damage detection based on hyperspectral technology.In this paper,fresh wire pepper planted at the pepper planting base of Guizhou Academy of Agricultural Sciences was used as the research object,and the data of field wire pepper was collected by hyperspectral data collection system,and the ripeness identification model and damage time prediction model of fresh pepper based on hyperspectral technology were established respectively to provide technical support for the automatic picking of pepper and nondestructive detection of pepper,and the main research contents and results are as follows.(1)The ripeness identification model of fresh pepper was constructed based on hyperspectral full-band data.Using the hyperspectral data of fresh pepper in the field,the normalized spectral data were pre-processed by SG smoothing,standard normal transform and first-order derivative method after reflectance correction and normalization,respectively.Then the SAM and SVM classification models based on hyperspectral full-band are established,and the six methods of SG-SAM,SG-SNV-SAM,SG-Der-SAM,SG-SVM,SGSNV-SVM,and SG-Der-SVM are compared and analyzed.The results showed that the field fresh pepper ripeness discrimination model based on SG-Der-SVM method had the most ideal recognition effect,and the F1 values of the recognition results for unripe fresh pepper,turning color fresh pepper,complete ripe fresh pepper,overripe fresh pepper and dried pepper were 100%,98.8%,100%,100% and 100%,respectively.(2)A ripeness recognition model for fresh peppers was established based on hyperspectral dimensionality reduction data.The normalized hyperspectral data of fresh pepper were used,and the hyperspectral data were downscaled using principal component analysis,and two neural network recognition models,BP and KELM,were combined to identify the spectral data of fresh pepper in the field before and after hyperspectral downscaling,respectively.The four methods of BP,PCA-BP,KELM and PCA-KELM were compared.The results showed that the overall effect of the field fresh pepper maturity discrimination model based on the full-band KELM method was the most satisfactory,with the F1 values of 100%,100%,100%,98.8% and 100% for unripe fresh pepper,trans-color fresh pepper,finish fresh pepper,overripe and dry pepper,respectively.(3)A prediction model of damage time of fresh pepper was simulated based on hyperspectral feature waveforms.The feature bands extracted by SPA and RF were used to predict the damage time of fresh pepper by combining two identification models,BP neural network and LSSVM.Finally,SPA-BP model and RF-LSSVM model were selected.For fresh pepper with damage time of Day0,Day2 and Day10,RF-LSSVM had the best recognition effect,with F1 values of 96.50%,96.50% and 96.04%,which were 9.48%,15.37%and 2.78% higher compared with SPA-BP model,respectively.For fresh peppers with damage time of Day4,Day6 and Day8,SPA-BP recognition was the best,with F1 values of100%,98.99% and 93.94%,which improved by 1.5%,4.15% and 2.25%,respectively,compared with the RF-LSSVM model;overall,in terms of recognition accuracy(Acc),RFLSSVM recognition was the best(Acc=95.5%).In this paper,we propose a method to extract pepper maturity and damage features based on the hyperspectral information of fresh pepper to solve the technical problems of fresh pepper maturity recognition and mechanical damage in natural environment.