Design And Experiment Of Feedstuff Type Identification System Based On Machine Vision

In the process of feed production,identifying the types of feedstuffs during the warehousing process is a key step.Which is mainly identified by manual sampling,relying on workers’ sensory experience.In order to realize online automatic identification of feedstuff types and improve the automation and intelligence of feed production,this study designed a multi-channel feedstuff automatic identification device and an online identification system.By establishing image data set of feedstuffs and using convolutional neural network method,developed a software suitable for the online identification system of feedstuffs,achieving online automatic control of feedstuff sampling and type identification,providing a new method and technical support for the automatic identification of feedstuffs during the warehousing process.The specific research content and results are as follows:(1)An online identification device for feedstuff was designed.The device has completed prototype processing,and the reliability and rationality of the structure of the device has been improved and optimised through testing.The device consists of a sampling unit,a material transport unit and a image acquisition unit.In which the online sampling unit is used to achieve online automatic acquisition of feedstuffs.The material transport unit is used to transfer the feedstuff samples that accumulated in the collection hopper to the image acquisition area.And the image acquisition unit can collect images of feedstuff samples reaching the bottom of the dark box in real time.(2)An online identification system for feedstuff was designed and developed.The Arduino Uno development board is used as the control core of the feedstuff oline identification device.The control process and control line were designed,and the control program was written and debugged in the development environment of Arduino IDE.Based on the Py Qt5 framework environment,the system software of the PC-based online identification system for feedstuffs was designed.The upper computer system software communicates with the lower machine controller through a serial port to enable the automatic control of the device.(3)The method of feedstuff types identification based on machine vision was studied.Images of feedstuffs were acquired using the feedstuffs types online identification device,after pre-process and perform data enhancement operations,a dataset containing 20,000 images was constructed.Four convolutional neural network ingredient identification models were established-Alex Net,Vgg Net16,Res Net18 and Res Net34.The identification accuracies of them are 95.80%,97.65%,98.80% and 98.70%,the Res Net18 model achieving the highest accuracy.The Res Net18 network model was optimized by adding the CAM channel attention mechanism and other optimization operations.The test identification accuracy of the improved Res Net18 network model reaches 99.40%.(4)The integration of system functions and system testing were completed.The improved CAM-Res Net18 network model was deployed to the online identification system for feedstuffs,and the PC operating system software with materials identification function was integrated.Achieve the intelligent operation of online sampling,image acquisition,type identification,result feedback and one-key alarm during the process.It was proved that the system hardware and software control systems run normally and meet the design requirements.The system performance test was carried out with the identification accuracy and identification time as the indicators of the whole machine performance test.The identification accuracy rate of peanut meal and wheat is 90%,and that of other feedstuffs is 100%.The overall identification accuracy rate is 98%,while the single sampling identification time of 10.13 s,which can meet the actual production requirements.This study designed and developed a feedstuff identification system suitable for practical feed processing production,which can be used to identify the types of feedstuffs online.The system achieves online sampling and automatic identification of feedstuffs during the warehousing process.The results indicate that the system have a good effect on the identification of feedstuff types and provide a new method and basic support of online automatic sampling and the identification for feedstuff types,which has important application value.