The Evaluation About Investigation Method Based On Computer Vision Of Rice Planthopper

Rice planthopper is a kind of important rice pest in China, it is a serious threat to rice production and food security in china when planthopper occured. The effective prevention based on accurately monitored and forecasted the occurrence of rice planthopper in dynamic. The investigation method of rice planthopper will inevitably affect the reliability of the study on the population dynamics of rice planthopper, quantity prediction, damage detection and control index. Now the investigation methods of rice planthopper are mainly disc plate making method (nothing and with viscose), visual observation and sticky board method. Plate making method is an extension census method of rice planthopper in the Plant Protection Station in China; plate making method (with viscose) is a method used in scientific research; sticky board method is applied on field investigation, and the visual observation method is a method which used before 70 in China plant protection station. The traditional method need high labor costs, the investigation personnel’s professional quality requirements, which make an irreconcilable contradictions between the requirement and lack of plant protection person in grassroots in our country, and the traditional methods cannot adapt to the scientific management of agricultural pests under the new situation. At present, the development trend of precision agriculture, large scale agriculture puts forward new requirements for plant protection work. The development of the computer technology, the machine vision technology provides a new methods for the scientific management of agricultural plant diseases and insect pests.This paper is mainly based on the identification and counting system of rice planthopper developed by China National Rice Research Institute, as there is a good recognition indoor static image, we make a comparative study of machine vision technology and traditional investigation method, so as to improvement and extension the rice planthopper investigation method based on machine vision. The results as follows:1) According to the number of rice planthopper on images, divided images into 5 level, statistics and analysis detection rate, false detection rate, loss rate of rice planthopper under 5 level images, at the same time, make the significance of analysis on the planthopper density between the actual number of the rice planthopper on the images and machine vision, find when the number of rice planthopper images is less than 2 head per clump, detection rate is 100%, but the error detection rate as high as 70.70%, t test results show that the program reads the results had significant difference with the actual amount of photos. The number of rice planthopper images is 2-8 head per clump,8-16 head per clump,16-30 head per clump, that the detection rate is 87.68%,85.59%,87.47%, and the loss rate 17.75%,15.40%,13.47%, t test result is0.286,0.566, 0.662,which is higher than 0.05,and show there are significant difference between the actual number on images and machine vision, in more than 30 head per cluster, the detection rate of 89.66%, false positive rate 20.21%, there is no significant difference between the program reads the result and the actual amounting. The intelligent system has good detection effect. 2) Preset the population of rice planthopper in rice plant in laboratory, then using different kindsof methods to survey the population during tillering stage of rice.The result shows when the rice planthopper density less than 2 head per cluster, the survey results of 4 kinds of traditional investigation method and machine vision method with the actual situation of the error is larger, the rice planthopper trapping rate would appear more than 100% of the amount of rice plant hopper, which means that the number of survey population are not reliable; when the preset density increased, the 4 traditional methods trapping rate increased, but the increase rate slowed down gradually, the trapping rate of machine recognition is reduced after the first decreased, in the 2-4 head per cluster> machine vision trapping rate is higher than that of the 4 traditional methods, and more than 4 head per cluster, machine vision trapping rate is lower than the 4 kinds of traditional methods. As the trapping population increased, the trapping rate has a stable change in machine vision and four kinds of traditional methods. Analysis the number of survey rice planthopper and the preset density, we find that using 4 kinds of traditional methods to get the number of survey insect has "S" growth trend as the preset density increased. And the machine vision has the similar change, so the machine vision is entirely feasible in practice, then fit the wheel plate making method, plate making method (with viscose), sticky board method and visual method and machine recognition and artificial photo counting (y) and density (x) seized the equationy=10.32/(1+e5.755-o.655x)、y=10.41/(1+e5.289-0.661x) y= 9.85/(1+e5-801-0.663x) y=8.65/(1+e6-808-0.662x)、y=6.15/(1+e3.202-0.751x)、y= 4.60/(1+e5.069-0.799x) As machine vision is feasible, than we set up the traditional method (x) and the method of Machine Vision Survey (y) between the conversion model.3) Using different survey methods to monitor the rice planthopper population density. At the tillering stage, using 4 kinds of traditional methods correspondence machine recognition to survey the population of rice planthopper in the field. Using the laboratory model to try to transform the popilation gotted by plate making method, plate making method (with viscose), the visual method, sticky board into the corresponding machine recognition result, and the conversion result the results of significant analysis and machine read at 95% confidence interval. In the TNI experiment area, rice plant hopper was relatively high, the results of analysis oft test were 0.082, 0.830,0.678,0.452, which are bigger than 0.05. In rice Xiushui 11 District, population density is relatively low, the results of analysis of t test were 0.868,0.890,0.136,0.608, which are bigger than 0.05, indicating that the model has a certain practicality. In the whole growth period of rice, different methods for dynamic monitoring of the occurrence of rice planthopper in rice planthopper population trends, found that the 4 traditional methods and technology of machine vision research in the monitoring period is roughly consistent. Therefore, the machine vision technology applied in field is feasible.