External Quality Detection Of Dried Strawberry And Software Development Based On Monocular Vision And Binocular Stereo Vision

Strawberry has special flavor and high nutrient value,but it is easy to rot during transport and storage.Dried strawberry is one of the strawberry deep-processing products which maintain its original flavor and nutrient contents to a great extent and reserve easily.It meets the demand of consumers and has important market position and broad development prospect.The quality of dried strawberry processed by vacuum freeze-drying technology is uneven,and there are great differences among individuals.When consumers choose dried strawberry,the first consideration is appearance quality.Computer vision technology can use a camera to collect images of fruit crisps,combined with image processing,feature extraction and other operations to achieve the detection of the external quality of the fruit crisps,such as surface defects,color,shape and size.Among them,the binocular stereo vision technology can locate the target and obtain its three-dimensional information.In recent years,it has been widely used in fruit identification and quality detection.Therefore,it is necessary to grade the appearance quality of dried strawberry.The traditional methods of fruit chips detection and classification are manual,time-consuming and labor-consuming,which is not suitable for the detection of production line.Therefore,the rapid nondestructive testing of fruit chips can not only enhance the productivity and quality of fruit chips,but also satisfy the consumers and promote the development of China’s fruit chips industry.Taking dried strawberry as the research object,monocular camera and binocular camera were used to obtain the image of dried strawberry,and the external quality feature parameters such as size,color,shape and texture were extracted.The traditional modeling method and deep learning method were used to build the model,and the best model was selected to develop the dried strawberry external quality detection and comprehensive classification software.This study can provide a theoretical support for further research of strawberry slices grading equipment on the product line.Jiangsu province’s key R&D project(Modern agriculture,project no.SBE2019310237)and Jiangsu province graduate practical innovation project(project no.2020)supported this research.The main research contents were as follows:1.External quality detection of dried strawberry based on monocular visionThe monocular vision system is built,which is mainly composed of industrial camera,industrial lens,bracket,light source,dark box and computer.Image acquisition and processing are respectively completed by self-written image acquisition software and Matlab.After a series of preprocessing,such as clipping,gray conversion and image segmentation,the pixel area parameters are extracted from the image as size features,the mean value of R,G,B and H color components as color features,the fruit shape index and the first 20 Fourier descriptors as shape features,and the contrast,correlation,energy and homogeneity as texture features.By inputting the above features into the traditional small sample classifiers SVM-C and PLS-DA,the SVM-C model is more effective in judging the external quality of dried strawberry.The classification accuracy of SVM-C model prediction set is 91.5%,and that of PLS-DA model prediction set is 83%.Using VGG16model of convolution neural network,based on transfer learning method,dried strawberry samples were learned and classified.The classification accuracy of VGG 16 model prediction set was 90.9%,higher than PLS-DA model,but slightly lower than SVM-C model.Therefore,the use of monocular vision technology to detect the appearance quality of dried strawberry can achieve better results and meet the requirements of grading accuracy.2.External quality detection of dried strawberry based on binocular stereo visionThe stereo matching algorithm of the binocular image was studied to calculate the three-dimensional information of the image,and the volume information was extracted from the binocular image as the size feature to reflect the appearance quality of the dried strawberry from a three-dimensional perspective.In the regression prediction model based on the volume of dried strawberry,the coefficient of determination(R~2)and root mean square error(RMSEP)of the regression model between the extracted volume and the measured volume were 0.9273 and 3.44 cm~3,respectively.The accuracy of SVM-C and PLS-DA was 91.6%and 86.6%,respectively.When using VGG16 model based on transfer learning to train samples,the classification accuracy of prediction set is 89.1%,which is better than PLS-DA model,but slightly worse than SVM-C model.The results show that the effect of using binocular stereo vision technology to classify the appearance quality of dried strawberry is better than monocular vision technology.3.Development of software for external quality detection and comprehensive grading of dried strawberryIn this study,based on the SVM-C model established by binocular images,the external quality detection and grading software system of strawberry crisp was developed using Matlab language under the Matlab R2016a platform.The software has a visual operation interface,which is simple and convenient.The software includes four modules:image import,quality detection,grade display and result saving.The volume prediction model based on binocular image of dried strawberry and SVM model were imported into the software,and then 120 dried strawberry were randomly selected for software validation.The results showed that the overall accuracy of dried strawberry appearance quality classification was 90.83%,which could meet the requirements of dried strawberry appearance quality detection and classification.