Construction Of Grading System And Study On Classification Of Fresh Lentinula Edodes Based On Machine Vision

Lentinula edodes is one of the agricultural products with Chinese characteristics,and it is also the most advantageous mushroom species in Hebei Province.Aiming at the industrial problems of high cost and low degree of automation of fresh Lentinula edodes classification,an automatic grading system of fresh Lentinula edodes is designed based on machine vision technology,which realizes the functions of automatic feeding,intermittent transportation,monomer,gesture reversal,grade classification and recognition and grading of fresh Lentinula edodes.The main contents are as follows:(1)The grading criterias of fresh Lentinula edodes were compared and analyzed.The four indexes of color,shape,cap diameter(size)and expanding grade of cap were determined to classify fresh Lentinula edodes.Fresh Lentinula edodes produced in different batches were sampled,and the physical character parameters such as the cap diameter,the stipe diameter,the whole length,the side stand height,the circularity and the sliding friction coefficient of samples were obtained by experiment.The statistical calculation results showed that the average moisture content of fresh Lentinula edodes was 83.40%,the average density was 0.80 g/cm3,and the average sliding friction coefficient between fresh Lentinula edodes and stainless steel was 0.51.There were significant differences among different grades of fresh Lentinula edodes,the higher the grade of fresh Lentinula edodes,the more round the cap.(2)The classification and recognition method of fresh Lentinula edodes based on convolutional neural network was studied.The images of five grades of fresh Lentinula edodes were augmented and preprocessed by flipping,rotation,translation,contrast transformation and brightness transformation,so the image data sets of the positive and negative of each grade fresh Lentinula edodes were established.Three kinds of pretrained deep convolution neural network models(AlexNet,GoogLeNet,ResNet-18)were used for transfer learning to quickly construct the classification and recognition model of fresh Lentinula edodes,called MoGu_Ale model,MoGu_Goo model,MoGu_Res-18 model respectively.Bayesian optimization algorithm,was used to determine the optimal combination of three hyparameters(Initial learn rate,Momentum,L2 regularization coefficient),then the respective optimal network model was trained.Through the test of the optimal network models,the accuracy of Z-MuGu_Res-18 model for classification and recognition of the positive images of fresh Lentinula edodes was the highest,reaching 98.73%,and the accuracy of F-MoGu_Res-18 model for classification and recognition of the negative images of fresh Lentinula edodes was the highest,reaching 99.15%.These two models could meet the classification requirements of fresh Lentinula edodes.(3)On the basis of physical character parameters,the overall structure design and the parameters design of the main components of the fresh Lentinula edodes grading system were carried out,including the design of vibration feeding device,monomer turntable flip device,grading conveying device and image acquisition device,so that the fresh Lentinula edodes could complete a series of actions such as automatic feeding,conveying,image acquisition,recognition and grading.(4)The motion simulation of the flipping gesture of fresh Lentinula edodes was carried out by ADAMS software.The motion track of the center of mass and the movement status of the turning process were obtained when the fresh Lentinula edodes was turned from the positive to the negative.The speed of the fresh Lentinula edodes was 70.122 mm/s when it was pushed on the monomer turntable by simulation optimization.The horizontal and vertical installation distance between the center of the monomer turntable and the center of the receiver were 184.77 mm and 70.111 mm respectively.The flipping platform was made and 30 experiments were carried out,and results showed that the average flipping success rate of fresh Lentinula edodes reached 90%,which proved the feasibility and accuracy of the design scheme in this study.