Study On Automatic Pest Counting And Identification System Based On Image

The agricultural pest damage in our country is very serious. There is also a great population of density and species. Spraying pesticide brings serious pollution for water and soil resources in reducing pests which resulting in a decline in quality of agricultural products and ecological system imbalance and other issues. Therefore, the plant monitoring and forecast are very important before the pests occur. The traditional manual monitoring methods which used artificial senses discriminate pests in the field and statistical quantity is influenced by man’s subjective factors and time-consuming. At present, with the development of intelligent detection technology continuously, the monitoring based on image has become the first choice.On the basis of the traditional trapping device, the photoelectric sensor counting method, the GPRS wireless communication technology, the android core control technology and pattern recognition technology were adopted. The pests automatic counting, wireless information transmission, information management technology were investigated. The moth was selected as the image recognition research object. Finally, a pest automatic counting and identification system based on the research was established.The main research work of this paper is as follows:(1)The overall design of the system was completed, a solar power lamp which can forecast the plant pests forwards according to actual condition was designed. Counters, infrared photoelectric sensor sensing camera module, GPRS module and the android control module were installed in the capture device. The accurate count of pests, the capture of pest image and remote wireless transmission of counting information and image information were realized.(2)A management information system was designed in PC, there is a GPRS receiving module in the front of the management information system. On one hand, it can send commands to the remote trapping device through the module, on the other hand it can receive real-time counting and image data from the remote capture device. When the management information system queried the information through the serial port, the received data can be saved to the database so that the information can be classified and integrated.(3)The moth image threshold segmentation was putted in HSV model, inflation in mathematical morphology, largest unicom methods were used in search and comparison to complete the image preprocessing, the foundation for the next step of feature extraction was laid.(4)On the basis of the study of feature extraction, the area, perimeter, eccentricity on the morphology and shape parameters, the complexity, duty ratio, maximum string, minimum string, ball and seven moment invariants20features were extracted as raw based on the analysis of the image, seven of them were choosed as the most effective feature set.(5)The BP neural network was studied, an improved classification method was choiced which is used for classification and recognition of insect pests, the improved classification algorithm and the original algorithm of learning efficiency were comparatived and analysised, the recognition of the moth was realized finally.