| Carrot(Daucus carota L.)is a kind of important economic and nutritional crop in China.Moisture content and carotenoid content are important indicators for evaluating the drying effect of carrots.It is necessary to find nondestructive detection methods instead of time-consuming,labor-intensive and destructive conventional detection methods to detect them.Combining the advantages of computer vision technology and spectral technology,hyperspectral imaging technology can be used for research on the external and internal quality detection of agricultural products.However,the high cost and huge data make it difficult to be applied to a real-time quality monitoring system directly.Therefore,hyperspectral imaging technology is mainly used to extract characteristic wavelengths in the quality detection of dried fruits and vegetables,in order to develop multispectral imaging systems suitable for portable or online detection.However,there are few studies on the establishment of multispectral imaging systems based on characteristic wavelengths and applying to quality detection of dried fruits and vegetables.Therefore,this study used the characteristic wavelengths extracted from hyperspectral experiments to realize multispectral imaging,and developed a portable multispectral detector for quality of dried fruits and vegetables to quickly detect the moisture and carotenoid content of carrot slices during hot air drying.This study is of great significance to achieve rapid and real-time detection of the quality of dried fruits and vegetables,improve the quality of them,and promote the development of drying industry.The main study contents and results are as follows:1.Detection of moisture and carotenoid content in dried carrot slices based on hyperspectral imagingTaking carrot slices during hot air drying as research materials,hyperspectral imaging systems were used to collect the spectral information of 400 to 1000 nm and 1000 to 2200 nm.The moisture and carotenoid content prediction models were established after that and their contents were visualized.Based on SPA,10 characteristic wavelengths from 400 to 1000 nm and 11 characteristic wavelengths from 1000 to 2200 nm of moisture content,as well as 12 characteristic wavelengths from 400 to 1000 nm and 11 characteristic wavelengths from 1000 to 2200 nm of carotenoid content can be extracted,respectively.By comparing different wavelength ranges and modeling methods,it was found that the optimal prediction models for moisture and carotenoid content are MSC-SVM models with 400 to 1000 nm.Prediction models based on full wavelengths(400 to 1000 nm)had R2p for 0.984 and 0.911,RMSEP for 0.380 g/g and 34.836 mg/100g,and RPD for 7.98 and 3.36,respectively.Prediction models based on characteristic wavelengths had R2p for 0.962 and 0.898,RMSEP for 0.612 g/g and 37.544 mg/100g,and RPD for 4.96 and 3.12,respectively.The models had good stability and strong prediction ability.In summary,hyperspectral imaging technology can realize the nondestructive detection of moisture and carotenoid content of carrot slices during drying process,and provides a reference basis for the establishment of multispectral imaging simulation equipment and the selection of filter wavelengths.2.Detection of moisture and carotenoid content in dried carrot slices based on multispectral imagingTaking carrot slices during hot air drying as research materials,the filter wavelengths were determined according to the characteristic wavelengths,and then a multispectral imaging simulation equipment was built.Based on SPA,the preferred wavelengths of moisture and carotenoids suitable for use in filters were further determined.The SVM prediction models of moisture content and carotenoid content were established based on the optimal wavelengths.The optimal prediction model of moisture content was established with 7 wavelengths,corresponding to 450,525,560,695,800,850 and 950 nm.The optimal prediction model of carotenoid content was established with 8 wavelengths,corresponding to 450,465,525,560,590,660,850 and 950 nm.The R2p of the models were 0.991 and 0.968,the RMSEP were 3.124%and 31.572 mg/100g,and the RPD were 10.32 and 5.34,respectively.The models had good stability and strong prediction ability.By comparing with the hyperspectral prediction results,the feasibility of detecting the moisture and carotenoid content based on SPA optimized wavelengths and the use of the multispectral imaging simulation equipment are proved.The detection of moisture and carotenoid content from hyperspectral imaging to multispectral imaging is portable and stable.3.Design and development of the portable multispectral detector for quality of dried fruits and vegetablesAccording to the requirements of portability and ensuring image quality,hardware devices such as development board,refrigerated camera,high-definition lens,bandpass filters,filter wheel,and display screen were selected.Through the design of software modules such as control module,data transmission module,calculation module and display module,the functions of information collection,transmission,processing and display were realized.Interactive interface guided users through operations such as model selection and quality detection.The sample was placed in the center of the sample compartment,and the camera and filter wheel were controlled by the development board to collect images of the sample under different spectral channels.After that,the model algorithm was called to obtain the quality information of the sample and output to the display.The image and quality information can be obtained directly through the local area network using a mobile phone or computer,so that the instrument can perform sample detection separately from the computer to achieve portability requirements.4.Modeling and verification of the portable multispectral detector for quality of dried fruits and vegetablesTaking carrot slices during hot air drying as research materials,SBS algorithm and SVM algorithm were combined to select the wavelengths of filters,and the portable multispectral detector for quality of dried fruits and vegetables was used to establish models for predicting moisture and carotenoid content.Detecting moisture content required 4 filters with corresponding wavelengths of 525,590,800 and 960 nm.Detecting carotenoid content required 7 filters with corresponding wavelengths of 450,525,560,660,695,850 and 960 nm.The R2p of the models were 0.989 and 0.920,the RMSEP were 3.668%and 29.803 mg/100g,the RPD were 9.15 and 3.50,while the average relative errors of the prediction set were 8.81%and 11.13%,respectively.The models had good stability and strong prediction ability.After importing the models into the instrument,the detection effect of the instrument was verified.The detection time was 45 s.The detection accuracy was high,while the average of the relative errors were 9.26%and 13.14%,respectively.The portable multispectral detector for quality of dried fruits and vegetables can be used to quickly detect the moisture and carotenoid content of carrot slices during drying process. |
PDF Full Text Request