Research On Method And Device Of External Quality Detection Of Dried Hami Jujube Based On Image Processing And Deep Learning

As a characteristic agricultural product in China,jujube is deeply loved by people for its sweet and soft taste and rich nutritional value.China has always been a big exporter of jujube,but the increase of added value and export value of jujube is not optimistic.One of the important reasons is that it is restricted by classification devices and detection technology.At present,the jujube grading device has a single detection index,and it is difficult to realize comprehensive external quality discrimination.Taking dried Hami jujube as the research object,firstly,normal,bird pecking,dry strip and mildew were detected by deep learning image classification.In order to solve the problems of high calculation,complexity and information loss of current residual network,an improved image classification method of deep residual network was proposed.Secondly,according to the grade difference of size and texture quantity,a threshold detection method was proposed,which can detect size and wrinkle by extracting the features of area,perimeter,fitting circle radius and texture quantity of dried Hami jujube image.The defect identification model,size and wrinkle detection model were integrated to realize the comprehensive detection of the external quality of dried Hami jujube.The main research contents and conclusions are as follows:(1)According to the actual inspection requirements of dried Hami jujube,the external quality inspection device of dried Hami jujube based on machine vision was designed,the testing platform was built,and the design scheme of key components was given.The image acquisition,transmission mode and light source configuration mode were determined.(2)In order to detect the size and wrinkles of dried Hami jujube,a threshold detection method based on image processing was proposed.Through image processing methods such as threshold segmentation,morphological processing,connected domain labeling and counting,the size features and fold features of dried Hami jujube were obtained.The size detection models of dried Hami jujube based on image area,image perimeter and fitting circle radius were constructed respectively.The results showed that the accuracy of size detection can reach 93.75% when the image area was taken as the size feature and the feature threshold was21230 pixels.When the texture quantity was taken as the wrinkle feature and the feature threshold was 37,the wrinkle detection accuracy rate reached 95.00%.The results showed that the size and wrinkle detection of dried Hami jujube can be effectively realized by image processing method.(3)In order to detect the external defects of dried Hami jujube,a defect detection method based on deep learning was proposed.By changing the input backbone of Res Net-50 network structure,7×7convolution kernels was replaced by three 3×3 convolution kernels,in which the step size of the first convolution kernel was 2,the number of output channels was 32,and the number of output channels of the last convolution kernel was 64,which greatly reduced the calculation cost and the number of parameters of the network model under the condition of ensuring the consistency with the original network output backbone information.In addition,by changing the number of channels of the downsampling module in the classification model,the input feature information was directly output after 1×1 convolution,which not only ensures the integrity of the input feature information,but also reduced the calculation cost.The research results showed that,compared with Res Net-50 network,classic deep convolution neural network model VGG-19 and Goog Le Net Inception v2,the model proposed in this thesis can converge faster under the same experimental conditions,and the accuracy of defect identification test reached 99.26%.(4)In order to verify the practicability of the external quality testing model of dried Hami jujube,the comprehensive testing model of external quality of dried Hami jujube was built by integrating the defect identification model and the size and wrinkle detection model,and a software for external quality testing of dried Hami jujube was developed.The functional areas of human-computer interaction interface were designed,which mainly included user login,real-time image display and processing result display,etc.,which provided users with a convenient operation environment and simple interface.The practicability of the system was verified by online detection.After testing,the detection accuracy was the highest when the conveying device speed was 0.5 m/s,and the overall detection accuracy was 92.50%.The detection efficiency was about120 pieces /min.The external quality inspection method of dried Hami jujube based on image processing and deep learning was proposed,which can effectively realize the size,wrinkles and external defects inspection of dried Hami jujube,and can initially meet the production requirements of online external quality inspection equipment of dried Hami jujube.The results provided theoretical basis and technical reference for online detection technology of fruit external quality.