Font Size: a A A

Design And Implementation Of Soybean Grain Appearance Quality Distinguish System Based On Random Forests

Posted on:2016-01-09Degree:MasterType:Thesis
Country:ChinaCandidate:R DingFull Text:PDF
GTID:2308330461498545Subject:Agricultural Electrification and Automation
Abstract/Summary:PDF Full Text Request
Our country is vast in territory and abundant in resource.Soybean planting has a long history in China. Soybean has a high nutritional and application value, having a close relationship with our lives. Known as the " king of beans ", " tianzhongzhi meat ".Quality-defect soybean is an important limiting factor of soybean production, estimated production losses amounted to 8.6%-12.3%.Only from the right and quick method distinguishing disease species, can make effective prevention methods to achieve the purpose of get twice the result with half the effort. The current detection means and identification disease on soybean appearance quality remained in the manual stage, hasing subjectivity, slowly, low accuracy of shortcomings. So using machine vision and pattern recognition technology to detect and identify diseased soybean species from the appearance quality is particularly important. The article studies identify soybean species(normal,gray-spot, worm-eaten, mildew, broken), building random forest classifier and designing the software part of the soybean appearance quality detection system based on Visual C++. The main research contents are as follows:(1) The article studies overall workflow of the soybean appearance quality detection system, complete whole design project that detects soybean diseases, and complete software system of overall planning and function modular decomposition process.(2) As a modern means, the pattern recognition technique can effectively improve the classification level of crops. With the development of technology, quick and accurate detection and classification of crop has become the main direction of the world. Random forest as an important method in pattern recognition, it has outstanding performance in regression and classification problems; characteristic parameters do not need to be selected and can deal with high dimensional; has strong processing ability for data containing noise and abnormal situation; its accuracy very high; the phenomenon of over fitting probability low. So been choosed to classify diseases in soybean identification in this paper. Then based on the idea of establishing more trees classifier to build a classifier into the forest for soybean identification and classification. Adoption point of comparison method as split guidelines, which is to compare differential values of pixels, without interference, so the method running time is short, with high precision.(3) The software system is an important part of running on the main processor detection system to complete the target of identification and real time display of work.To study efficient software technology applied to display the development process of the platform is an important aspect of improving system function, also can accumulate the technology for the development of similar software. The software system is developed in Visual C++ environment, which has a friendly interface and strong, low maintenance cost, easy operation, low complexity, low power consumption advantages and user convenient for daily using. After analyzing the detection system of the main work, the software system is divided into user login module, image feature extraction module, image processing module, image file operation module, soybean classification module, historical data query module and data storage module, and introduces the main use of software technology, which on the PC interface display effect is good.In recognition system to select 10 kinds of soybean samples that cone from northeast agricultural university institute of soybean, including insect normal, gray-spot, mildew, broken, worm-eaten. Random forest classifier for four kinds of defect of soybean recognition rate reached 94%, 96%, 92% and 91%, the mixed soy sample recognition rate reached 81%, to achieve the desired effect. For the development of software system, through indoor test, system functions run normally. Results show that: the communication normal, real-time and effective parameter display, in the process of display screen flicker phenomenon does not occur, meet the basic requirements of the system.
Keywords/Search Tags:soybean, distinguish quality, Visual C++, pattern recognition, random forests
PDF Full Text Request
Related items