The Research On Intelligent Identification And Counting Technology Of Rice Light-trap Pests Based On Images

At present,pest monitoring methods in China include insect radar,light trap and field investigation.Rice insects are mainly induced and killed by pest-killing lamps using a specific spectral range,collected by the researchers in the next day.The statistical data acquired by artificial identification and observation will be used to monitor the dynamics of the reported population of pests.This artificial identification and calculation methods have problem of low-efficiency,poor instantaneity and high labor intensity.On the basis of He Xinjiang,the thesis carry on the research of image-based rice light-trap pests identification of insect pests reported by rice light trap.The main research contents,research achievements and innovations of this thesis are as follows:(1)Using intelligent insect detection light trap to collect and preprocess image of insect pest in real-time.Firstly,performing distortion correction,background removal and morphological processing and other operations to the collected images.Then,acquiring connected regions in the collected images and cutting the minimum enclosing rectangle to obtain a single image of rice pest entrapment.Finally,dividing the rice light-induced pests into large-sized pests and small-sized pests according to the pixel amount in the connected area of a single rice lamp-induced insect image and the length and width of the minimum circumscribed rectangle.(2)Studying the effect of different image features on the results of rice pest image recognition.In the traditional pattern recognition methods,the selection and fusion of different image feature parameters have a significant impact on the recognition results.In this thesis,by study the effects of global,local and fusion of different features on the identification of target pests in rice insect images,and obtain the support vector machine classifier based on global features combine HOG features fusion training and the support vector machine based on global features.And we can get more ideal results through respective identification and classification of large-size pests and small-sized pests.(3)Studying the impact of the ratio of different target samples and non-target samples on the results of rice pest image recognition.Identifying the target object from a large number of interfering objects,the proportion of different training samples has a greater impact on the recognition results of the target object.Under the condition of small sample of targeted rice pests,the research choose different ratio of training samples to test the identification of target pests,when the non-target sample size was about 4 times of the target sample size,three kinds of large rice light-trap insects access to 91.4% identification rate and a 8.6% false positive rate.When the non-target sample size is about 2 times of the target sample size,two kinds of small rice light-trap insects can get 94.9% identification rate and 4.9% false detection rate.