Study Of Classification Of Fruit Flies Insects Based On Digital Image Processing Technology

In this paper, digital image processing and pattern recognition technologies are applied to identify 27 species of pest fruit flies in Yunnan. These pest flies belong to three subfamilies, four families and eight genus. Auto-identify different species is realized in this paper. The main content and achievement as follow:(1) Pre-processing of 27 species of pest fruit flies image. This paper mainly studied the processing of color image binarization and image denoising. The results showed that the blue component has a greater contrast, and image detail is also more clearly. Median filter is used in blue component image, then noise interferences are reduced.(2) Segmentation of 27 species of pest fruit flies image. Using adaptive threshold algorithm, a fly image are divided into five parts:head, chest,abdomen, left wing and right wing. We take use of different segment algorithms for different parts of fly. Improve operation of removing spurs in traditional morphological algorithms, and applied to segment transparent wings. Experimental results show that good effect of segmenting fly image may be got under the condition of strong noise background.(3) Feature extraction and feature selection of 27 species of pest fruit flies image. Analysing the characteristic of different parts of fly image and take the large image noise into account, we extract area,markings,weight, color,texture in wing and chest image and many other features. The results show that these features are typical and effective, they can be extract in image under a strong noise and don't change with the image shift, rotation or scaling.Classification of 27 species of pest fruit flies image. Because of the limited sample of fly images, we conbined the binary tree and support vector machine classification method in order to improve real-time processing capability and to reduce the ratio of misclassification. When features are sharply different, binary tree can make processing of classification faster and more clear. Support vector machines is a specialized study of a small training sample cases in comparison with the traditional statistical theory. It solves the problem of over-study and local extremum in Neural Network Algorithm. Experimental results show that, using the method of combineing a binary tree and support vector machine to classify is effective.