Research Of Intelligent Sorting System For Cordyceps Sinensis Based On Deep Learning

Cordyceps sinensis as a kind of traditional Chinese herbal medicine has been Prevailing for many years because of its high officinal value.Nevertheless,the terrible condition of growth which leads to the difficulties of picking up,results in a high price in the market.Recently,with the improvement of living standards,there is an increasing tendency for demand of cordyceps sinensis,which give rise to the problem of fake one here and there.And it will end in great financial loss to acquirers without doubt.Thus,this thesis will center on the intelligent sorting system for cordyceps sinensis aiming at solving the problem about pinch of cordyceps sinensis picking equipment,low efficiency of manual work and high error rate with the combination of image processing and flush type.The system combines image processing technology and embedded technology to identify the Cordyceps and its fake images through deep learning algorithm and complete the sorting work,especially on the Cordyceps recognition algorithm.The multi-point feature image extracted by SURF algorithm is used as the convolutional neural network input.The artificial bee colony algorithm replaces the traditional gradient descent method to optimize the weight update speed.Compared with the traditional algorithm,the recognition accuracy of the system is greatly improved and the recognition speed is accelerated.The main work of this thesis can be summarized as follows:(1)Project design and love technology analyzation of cordyceps sinensis sorting and distinguishing.Firstly,this thesis will commence on three aspects,development and applications on different areas of deep learning,comparison of main model to research and get a primary framework.Secondly,structure the flow path on the basis of demanding function and design philosophy of system,image recognition and acquisition which includes processing and identification and sorting control make up the system.(2)Hardware design and analysis of the system.Structure a program about holistic hardware construction according to analyzation of system's demand.First,finish column about image acquisition by designing industrial camera,camera lens,light source,image capture card and infrared detection.Second,control module is built by the selection and design of MVP,minimum system,steering gear and display screen to realize the function of control sorting.(3)Analysis and design of system software.Configure the parameters of the image acquisition card in the image acquisition module and external trigger function of the industrial camera can complete trigger photographing when the object enters the image recognition area.Then,OpenCV visual library is linked to anaconda platform of upper computer,python language and keras framework to design the basic image recognition process,which pares the way for the next image recognition algorithm and designs upper computer functions.Finally,through the software design of infrared detection sensor,steering gear and display screen,the function of cordyceps authenticity sorting is realized.(4)Research on image preprocessing and image recognition algorithm.Research on Image Preprocessing and Image Recognition Algorithm.In the preprocessing stage,image noise is processed through image graying,Gaussian filtering and median filtering,image enhancement is performed through histogram equalization,and Canny performs image edge detection to ensure that the image input quality approaches the ideal result.Finally,an improved convolution neural network algorithm S-CNN is proposed.The SURF algorithm is used to extract the image as the input layer of the neural network,and the artificial light bee colony algorithm is used to update the weights.The experimental results show that the recognition accuracy of the improved algorithm is 95.58%,and the recognition speed is increased by 0.23s/ unit.In this system,an industrial camera,a lens,a light source and an image acquisition card are built,and software parameters are configured to realize an external trigger function of the software,so that the image is triggered to take a picture through an infrared detection module and uploaded to an upper computer after the image enters an acquisition area.An Anaconda software platform is built on the upper computer to link OpenCV visual library,a Keras framework is built through Python language,and the identification of genuine and fake cordyceps images is completed through convolution neural network algorithm and transmitted to the sorting control module of the lower computer.The lower computer selects STM32 microprocessor to receive the upper computer information,judges and controls the steering gear sorting according to the infrared detection sensor,and displays the sorting information.At the same time,the paper proposes a convolution neural network algorithm of S-CNN.The experimental verification shows that the algorithm can identify the authenticity of Cordyceps sinensis,and has high identification efficiency and fast identification speed.