Machine Vision Inseption Of Wheat Apprearance Quality Based On Deep Learning

Wheat is the main grain crop and important commodity grain and reserve grain in our country.In the production,circulation and consumption of grain,the detection of wheat appearance quality plays an important role in the classification and grading of wheat quality.At present,most of the grain detection experiments are still in the stage of artificial naked eye observation.Aiming at the problem that image features need to be extracted artificially and recognition rate is not high based on machine vision technology,this paper builds the detection target of wheat grains,which are perfect grains,imperfect grains(damaged grains,diseased grains,germinating grains,mildew grains,insect erosion grains)and impurities(wheat husk).The wheat grain image acquisition platform of machine vision system has studied image preprocessing and segmentation methods.The traditional SVM(Support Vector Machine)recognition method based on artificial extraction features has been studied.The method of building recognition model based on deep learning has been studied in detail.The main contents and conclusions of this paper are as follows:(1)In order to solve the problem that most of the existing studies use manual display of Wheat Images under laboratory environment,a machine vision wheat grain image detection platform is designed and developed,which consists of inlet and outlet,orifice plate,brush,brush driving device,industrial camera and light source.The wheat grain is thrown into the orifice plate by the inlet port,and the brush located above the orifice plate moves in a straight line driven by its driving device,which makes the wheat grain evenly spread in the orifice of the orifice plate and clears the surplus grain into the grain recovery box.The image of the wheat grain in the orifice plate and the orifice is obtained by industrial camera and transmitted to the computer for processing and analysis.The collected wheat grain images are clear and non-sticky,which lays a good foundation for subsequent image preprocessing.(2)Aiming at the problem that the collected batch wheat image needs complex image preprocessing algorithm to remove noise interference and effective segmentation algorithm to segment the single-grain image,this study will collect the original image through gray processing,median filtering,threshold segmentation,morphological processing and other operations,making the original image remove noise interference,image.The quality is enhanced,and the single grain target is segmented quickly from the original image,which provides a good basis for the subsequent artificial feature extraction parameters.The single grain image segmented by the minimum outer rectangle method has different sizes.For the convenience of subsequent experiments,all other samples are expanded on the basis of the largest sample in the segmented size.Make it the same size as the reference sample.At the same time,the location of wheat grains in the hole is random.By rotating and flipping the single grain image,each sample is expanded to 4,and the overall sample size is expanded to 4 times of the original,which provides more and more comprehensive samples for the followup in-depth learning model training.In order to solve the problem that most of the existing studies use manual display of Wheat Images under laboratory environment,a machine vision wheat grain image detection platform is designed and developed,which consists of inlet and outlet,orifice plate,brush,brush driving device,industrial camera and light source.The wheat grain is thrown into the orifice plate by the inlet port,and the brush located above the orifice plate moves in a straight line driven by its driving device,which makes the wheat grain evenly spread in the orifice of the orifice plate and clears the surplus grain into the grain recovery box.The image of the wheat grain in the orifice plate and the orifice is obtained by industrial camera and transmitted to the computer for processing and analysis.The collected wheat grain images are clear and non-sticky,which lays a good foundation for subsequent image preprocessing.(3)Aiming at the problem that traditional manual feature extraction needs a lot of feature parameters and is easy to cause data redundancy,this study extracts 12 color features,10 morphological features and 5 texture features from segmented single-grain images,and processes 28 effective features by principal component analysis,and obtains the optimal 8 effective features as input parameters of the model.And it was input into the established traditional SVM model.The correct recognition rate of seven types of wheat grains was 80.2%.Among them,the recognition rate of perfect grains was the highest(86.3%)and the recognition rate of impurity wheat hull was the lowest(72.5%).The experimental results show that the traditional artificial feature extraction and feature optimization input to SVM model are feasible for multi-classification of seven wheat types,but the recognition accuracy and time are long.(4)Aiming at the problem of low recognition results of traditional SVM model,a method of wheat perfect grain,imperfect grain and impurity recognition based on deep learning model is proposed.In this method,ReLU function is used instead of traditional Sigmoid function in classical CNN(Conventional Neural Networks)network model to reduce the convergence time of the model.Dropout technology is introduced in the whole connection layer to reduce the over-fitting shadow.On the basis of conventional CNN network model,residual block structure and Bach Normalization algorithm are introduced.At the same time,the core size of the first convolution layer is changed to 5*5 and the dotted residual block structure is omitted.ResNet network model is constructed.The results showed that the correct recognition rates of the classical CNN model and ResNet model were 90.6% and 96.3% respectively for seven types of wheat grains.ResNet model has the highest recognition accuracy,which is 5.7% and 16.1% higher than classical CNN model and SVM model,respectively.The testing time of each group is 56 seconds,which can meet the requirement of rapid quality testing for wheat purchase in grain depots.