| Waxberry(Myrica rubra cv.Dong Kui Orient Pearl)is an important economic fruit with high nutritional value in Zhejiang Province.Waxberry fruit has no epidermal protection,makes the picking,inspection,transportation and storage of fruit more demanding,otherwise it will affect its quality and reducing economic benefits.At present,the quality detection method of waxberry is manual detection,which is inefficient and will cause accumulation and decay during the detection process.Therefore,a rapid nondestructive detection method is needed to accurately and quickly detect and classify the quality(size,sugar content,acidity).Based on the problems,this paper establishes the quality(size,sugar content,acidity)detection model and verification of waxberry based on binocular vision images and hyperspectral images through systematic experiments.However,hyperspectral detection is not suitable for waxberry industry because of its high cost.In response to this problem,a low-cost method for hyperspectral image acquisition for waxberry quality detection is proposed.The quality detection of Myrica rubra can be completed by transforming RGB image of sugar and acidity feature band into spectral image by synchronous correction method,and the detection method data acquisition sensor is reduced from hyperspectral sensor to RGB image sensor,which greatly reduces the detection cost.Finally,based on the above method,a low-cost portable waxberry quality detector was developed,which can detect the sugar and acidity of waxberry fruit and complete the quality detection.The experiments and research in this article mainly include the following aspects:(1)Carry out systematic experiments to obtain research data.The experimental data used for detection model and synchronous calibration are obtained.Waxberry detection model experiments include:(1)Image detection experiment of simulating human eyes.The purpose is to establish a waxberry detection model that simulates human eyes through RGB images,and collects 6000 images of waxberry with different quality;(2)Binocular image acquisition experiment of waxberry based on binocular vision.The purpose is to measure the quality parameter-size of waxberry through binocular images,and collects binocular images of 200 waxberry samples;The hyperspectral detection model construction experiment,aims to select the related components and sensitive spectral features of components,such as sugar content and acidity,and to establish a fitting model between spectral features and components..The experiment includes:(3)The spectral image collection of waxberry,a total of 200sample spectral images are collected;(4)Determination of chemical components related to the quality of waxberry,the contents of anthocyanin,reducing sugar,total sugar and PH value of 200 waxberry samples were determined by chemical methods.Low-cost method research experiments include:(1)Synchronous calibration model data acquisition experiment,aims to establish a synchronous calibration model of non-imaging/imaging spectrum,and collect non-imaging/imaging spectral incident radiance data and corresponding synchronized illuminance values through the whiteboard and illuminance meter,wherein the illumination intensity range is20000-100000lux,including the illumination intensity range of normal sunny weather;(2)Synchronous calibration model verification experiment,aims to verify the application accuracy of the synchronous calibration model,and collect the non-imaging/imaging reflection radiance and standard spectrum of vegetation,water and soil.These experiments provide a data basis for the subsequent construction of waxberry quality detection models and low-cost research.(2)Study on quality detection mechanism of waxberry.Including image vision and hyperspectral quality detection of waxberry quality.Waxberry quality detection with image vision is based on experiment(1)to collect waxberry RGB images,and carry out target recognition and grade classification of waxberry by deep learning of Faster R-CNN algorithm.The result: Waxberry fruit recognition has high accuracy under different environmental conditions,but the accuracy of grade classification results is low.The hyperspectral waxberry quality detection includes two parts:waxberry size,sugar content and acidity detection.The size detection is based on the binocular vision algorithm to process the waxberry binocular image in experiment(2).The waxberry size is detected through the processed depth map,which is ideal The verification result is that the RMSE is 0.369 and the RRMSE is 11.24%;the sugar acidity test is based on the data collected in experiments(3)and(4).Through correlation analysis,reducing sugar and p H are selected as the relevant chemical components of sugar content and acidity,and the corresponding The spectral characteristics are in the 610 nm and 570 nm bands,and the detection model of the characteristic band reflectance,reducing sugar content and PH value is further established,which has ideal results.The model verifies that the RMSE is 1.399 and0.1329 respectively.(3)Study on low cost of waxberry quality detection method.In this part,the RGB image of the characteristic band is converted into the spectral reflectance image by reducing the spectral resolution of the characteristic band and synchronous correction,so as to achieve the purpose of low cost of bayberry quality detection,which is verified by the RGB camera system equipped with the characteristic band filter.In the study of spectral resolution reduction of characteristic bands,the 4nm characteristic band of the sugar acidity in the detection model and the 32 nm band centered on this band were analyzed,with RMSE of 0.0023 and 0.0021 respectively.The RMSE of reflectivity for further detection of sugar content and acidity was 0.04 and 0.001,respectively,indicating that the spectral resolution of the characteristic band can be reduced from 4nm to 32 nm.In the synchronous calibration study,a fitting model between the incident radiance and the synchronous illuminance of each band of the non-imaging/imaging spectrometer was established,R2 was between 0.95-0.99,RMSE of the test set was within 0.035,which was verified by three objects.It is verified that the relative error of each band of non-imaging spectrum is within 5%,the relative error of imaging spectrum 450-720 nm is within 5%,and the relative error of720-900 nm is within 10%-25%.On the basis of the above research,two industrial cameras equipped with two feature wavelength filters of sugar content and acidity were used to collect grayscale images in the characteristic band of 30 nm respectively,and the synchronous calibration models of the two cameras were established respectively.The model R2 is 0.98 and respectively.0.99.Based on the camera synchronous calibration model,the waxberry characteristic band spectral image was reconstructed,and the test results of sugar acidity were verified.The verification accuracy RRMSE of sugar content and acidity were 14.29% and 4.64%,respectively.(4)Development of low-cost waxberry fruit quality detector.Based on the above research,a low-cost waxberry quality detection instrument was developed.Different from the characteristic band information collection method of the filter + natural light source in(3),this instrument collects by controlling the wavelength of the light source.Two LED lights are selected as the light source in the instrument,and the wavelengths witch have banwidth of 30 nm,are the sensitive bands of sugar content and acidity respectively.Illumination data collection The illumination information is collected through the illuminance module,and the characteristic wavelength reflectance and sugar acidity of waxberry are obtained through the synchronous calibration model and sugar acidity detection model in the chip to complete waxberry quality detection.This paper mainly studies the waxberry non-destructive quality detection method based on hyperspectral,and further reduces the cost of the detection method,and develops a low-cost waxberry quality detector,which realizes the function of low-cost and rapid non-destructive detection of waxberry quality,while taking into account the high With the high-precision feature of spectral detection,the detection cost is reduced from hundreds of thousands to thousands,which has a large application space. |
PDF Full Text Request