Research On Classification Of Main Insect Pests Of Noctuidae Based On Image Recognition

China is the world's largest tobacco producer and consumer.35% of global tobacco production and 32% of tobacco sales are in China.The tobacco industry is developing rapidly,and the economic benefits are also increasing year by year.Over trillions of taxpayers have been paid in succession for many years.Leading the industry and becoming the pillar industry of the national financial revenue.However,because of pests and diseases,it has brought enormous economic losses to the country every year.The damage caused by two kinds of pests,Helicoverpa armigera and Helicoverpa assulta,is very serious.The two coexist steadily only on tobacco,which are difficult to distinguish under human eyes.According to the characteristics of two kinds of insect adults and pupae,the automatic recognition technology of tobacco pests is studied by using image processing and pattern recognition techniques.The modern recognition method can quickly and accurately detect and identify pests to be tested.It is beneficial to the informatization and modernization of tobacco industry in China.The main contents of this paper conclude four parts:(1)Image collection and preprocessing of tobacco pests.Through the indoor breeding of two kinds of tobacco pests,using industrial camera to collect adult images of cotton bollworm and oriental tobacco budworm,in order to improve the contrast of the original image and strengthen the edge and other details are used to histogram equalization,mean filtering,etc.Pupae images were collected by industrial camera and SLR camera then determine the best acquisition equipment.(2)Segmentation of tobacco pest images.The RGB color space and HSI color space are transformed respectively.The results show that the adult image is more suitable for segmentation in S-channel.Otsu method and histogram threshold method are used to segment the image respectively.The results show that histogram threshold method is the best method for segmentation.Aiming at the noise and internal holes of the target image,morphological processing and hole filling are used to obtain a single target image,which is conducive to image feature extraction.The pupa R-channel image has good effect and can be directly used for feature extraction in the next step.(3)Feature extraction of tobacco pest image.By studying the color,texture and morphology of adult images of cotton bollworm and oriental tobacco budworm,we extract color moment features from RGB image of adult,texture features based on gray level cooccurrence matrix and difference statistical matrix and seven-order invariant rectangular state features from B-channel image of adult,and form a 36-dimensional image feature space.The difference of texture between pupae images of two pests was analyzed,and the texture features of R channel based on gray level co-occurrence matrix were extracted to form 16-dimensional image feature space.(4)Feature optimization and classification of tobacco pest image.The original feature space of adult image is optimized by simulated annealing algorithm.Then selected 15 optimization features such as B-channel second-order moment and G-channel third-order moment.Support vector machine is used as the final classification tool.There were 400 adult images in the sample,of which 280 were used for training and 120 for testing,and 280 images for pupal period,of which 200 were for training and 80 for testing.The recognition rate of two classes and four kinds of adults was 95.83%,2.5% higher than that before optimization,and time efficiency increased by 32.47%.The recognition rates of male and female pupae of cotton bollworm and oriental tobacco budworm were 82.5% and 87.5% respectively.The experimental results show that the image recognition technology is effective for adult classification and male and female discrimination of tobacco pests,and it is feasible for male and female classification of pupae stage.