Fruit Fly Insect Image Feature Extraction And Recognition Technology Research

Tephritidae are herbivorous insects which is widely distributed throughout the world,it mainly feeds on fruit and vegetable crops.It have been reported as an important quarantine for its great impact on agricultural production. Yunnan,which bordered with many countries in South east Asia, is an important origin producting fruits and vegetables.Its possibility of invasive species is very large,therefore,effective detection and quarantine of Tephritidae is necessary.Image-based identification of harmful insects is a hotspot in research of harmful insect detection,compared with human quarantine, it could save a lot of manpower and resources.In this paper,we use the 27 kinds of harmful Tephritidae of Yunnan Province as our research object.the pretreatment,image segmentation,feature extraction and classification techniques were studied in our research.First,pre-processing and image segmentation were done for the need of feature extraction.Image denoising,gray level transformation,edge detection and thresholding methods were used for processing of the image and edge images,speckle images and thresholded images were geted for post-processing. In order to extract more effective color features,RGB color space were converted to HSI color space which is more suitable for the study of color. Combined the knowlege of mathematical morphology, we separate the imsect body parts from the image,and the region for extraction of local feature were getted. Secondly,color feature,texture feature and morphology feature were extracted from the processed images and the original feature space consisted of them.Then SVM-based feature selection were done for reducing the dimension of morphological feature,and PCA-based feature extraction were done for reducing the dimension of color feature.Then the results of the dimension reduction were compared with the variance method and quantification method. finally,we get the feature space of dimension reduction as an input into the classifer,and the results of classification were recorded.In the classifier design,we chose the support vector machine classifier,evaluation and analysis were done through our classification results.