Main Quality Detection Methods Of Malus Micromalus Makino Based On Machine Vision And Spectral Analysis

Haihongguo（Malus micromalus Makino）is a unique economic fruit in China,which has high nutritional and economic value.The development of fruit planting industry is an available way to boost the wealth of growers,stimulate local GDP,and achieve rural revitalization strategy.Non-destructive,rapid and accurate completion of the main internal quality inspection and appearance quality evaluation of Malus micromalus Makino,and real-time automatic classification,are of great significance to improve its post-harvest commercial processing efficiency and brand premium capacity.However,there are some problems,such as the lack of nondestructive testing models of main internal quality,the difficulty in testing the external quality characteristics,and the lack of special automatic testing and grading systems.Aiming at improving the automatic processing level and production efficiency of post-harvest detection and grading of Malus micromalus Makino,the following work has been carried out respectively.Firstly,the nondestructive prediction model of main internal quality（soluble solids content,firmness index and maturity）of Malus micromalus Makino based on spectral analysis technology was studied.Then,the method of external multi-surface image acquisition and external quality classification based on machine vision technology were studied.Finally,an intelligent online grading system for the quality of Malus micromalus Makino was developed.The main research contents and conclusions of this paper are as follows.（1）Aiming at the lack of nondestructive testing models for the main internal quality of Malus micromalus Makino,a prediction and estimation model of near infrared spectroscopy for rapid detection of soluble solids content（SSC）was established.In this paper,9pretreatment methods,such as moving average,median filtering,normalization,baseline,de-trend and first derivative of direct differentiation,were compared.Then,the influence of3 data dimensionality reduction methods,such as random frog（RF）,successive projections algorithm（SPA）,and principal component analysis（PCA）,on the model is analyzed,and 12component prediction models were finally established.The experimental results showed that it is feasible to invert the SSC of Malus micromalus Makino by using the near-infrared spectroscopy in the range of 400-1083 nm,and the SPA-LS-SVR model has the best performance with the values of the calibration correlation coefficient(R_cal),the prediction correlation coefficient(R_pre),the root mean square error of calibration（RMSEC）and the root mean square error of prediction（RMSEP）of 0.962,0.902,0.199 and 0.271,respectively.（2）A hyperspectral imaging-based method for internal quality estimation and maturity stage discrimination of Malus micromalus Makino was proposed.In this paper,the distinctive wavelengths responsive to SSC and FI variations were extracted using the SPA,interval random frog（IRF）,and competitive adaptive reweighted sampling（CARS）algorithms.These wavelengths were used as model inputs to establish the internal quality prediction model based on partial least squares regression（PLSR）and extreme learning machine（ELM）,and to build the maturity classification model based on support vector machine improved by grey wolf optimizer（GWO-SVM）and partial least-squares discrimination analysis（PLS-DA）.The experimental results showed that it is scientific and feasible to quantitatively and qualitatively analyze the SSC,FI and maturity level of Malus micromalus Makino by using the near-infrared hyperspectral imaging technology in the range of 393-1016 nm;SPA is superior to IRF and CARS in the extraction of effective spectral bands.Moreover,only by relying on 0.03%band information,the SPA-ELM model can achieve the optimal SSC prediction results with the values of R_preand RMSEP are0.941 and 0.634,respectively;And the best FI estimation results with the values of R_preand RMSEP are 0.957 and 0.509,respectively.In the classification model of mature stage,SPA-GWO-SVM and SPA-PLS-DA have the best result,and the classification accuracy of both test set and independent verification set is more than 97.0%.（3）Aiming at the difficulty of appearance quality feature detection,a multi-surface image information acquisition device was developed,and a fast image damage feature extraction method was studied.Focusing on the characteristics of"small fruit diameter,quasi-sphere,bright surface",a chain transmission module with a micro-pitch unpowered dumbbell roller chain as the core structure,a flexible buffer friction transmission structure and a binocular vision multi-surface image information acquisition device were designed,which realized the complete acquisition of multi-surface images of Malus micromalus Makino.On this basis,the key technologies of image processing were designed,such as the image enhancement algorithm based on HSI color model and low-pass filter,the adaptive threshold multi-target segmentation algorithm based on Otsu which was optimized by particle swarm optimization（PSO）strategy,and the dynamic access and quick stitching strategy of multiple images.The experimental results showed that the above method effectively eliminated the complex background and noise redundancy in the original image and realized the rapid detection of the main appearance quality features of Malus micromalus Makino.（4）To solve the problem of the lack of automatic testing and grading system for the main internal and external quality of Malus micromalus Makino,a nondestructive testing and grading system for the main quality of Malus micromalus Makino based on spectral analysis and machine vision technology was developed.Firstly,the influence of the spatial distribution of spectrometers and light sources on SSC prediction model was studied,and the quality grading model of random forest was optimized based on sparrow search strategy.Then,the hardware of quality inspection system of Malus micromalus Makino with multi-surface image acquisition module and spectrum acquisition module as the core structure was developed by adopting the design idea of"building blocks and modularization",and the quality grading software of Malus micromalus Makino was developed by relying on C#and Halcon platform.The experimental results of the system showed that the developed hardware device can completely acquire the multi-surface images and spectral information of fruits,and the developed software can effectively complete image target detection,segmentation,synthesis,appearance damage feature extraction and automatic quality grade classification.Moreover,the classification efficiency of the whole system is 30 per minute,and the classification accuracy rate is 87.5%,which meets the design requirements of the automatic detection and classification system of Malus micromalus Makino,and provides a strong technical support for post-harvest commercialization of fruits.