【作者】 林冬瑾；

【导师】 黄星奕；

【作者基本信息】 江苏大学 ， 食品科学， 2007， 硕士

【摘要】 鞘翅目害虫是重要的农业和林业害虫，目前鞘翅目害虫的鉴定工作是由数量有限的专家进行，大部分非专业人员因缺乏相关知识的技术培训，要快速鉴定害虫种属具有相当的难度，不利于害虫防治。为此，本研究利用计算机视觉技术对34种重要的农业和林业鞘翅目害虫的自动鉴定技术进行了研究，设计了一个用于鞘翅目害虫自动鉴定的计算机识别系统。针对不同种属的鞘翅目害虫体长差异较大这一问题，采用了不同的图像获取系统分别采集体长不同的鞘翅目害虫图像：采用自行设计的鞘翅目害虫图像获取系统采集常规害虫(体长大于1cm的害虫)图像；而对于微小害虫(一般体长2-14mm的小型害虫)则采用带有显微镜接口的奥林巴斯数码相机连接显微镜进行拍摄。将获得的害虫图像进行图像增强、去噪声等预处理操作后，为了有效地分割出害虫，对不同种类的害虫图像提出了不同的分割方法：对微小害虫采用基于固定取阈法的分割方法来分割前景，对于常规害虫采用基于最大类间自动取阈法的分割方法进行去背景操作。针对去背景后的图像中包含的害虫触角和足节等部分，提出了稳定而有效的去噪方法。根据鞘翅目害虫的分类学特征，从害虫图像中提取了区域周长、区域面积、主弦，次主弦，离散指数，偏心率和紧凑性等七个形状特征参数作为鞘翅目害虫的分类判别依据。为提高鞘翅目害虫的识别正确率，研究了基于极大极小准则的模糊模式识别方法，基于多元分类的最小二乘支持向量机方法和BP神经网络方法，并将最符合实际分类要求的模糊模式识别方法确定为鞘翅目害虫的最终分类方法，利用该方法对34种鞘翅目进行自动鉴定时平均准确率可达92.16％。利用VB和MATLAB开发了鞘翅目害虫图像自动鉴定软件系统。该系统界面友好，功能齐全，使得只具有初步相关经验的用户利用该系统就可以进行专家水平的鉴定工作，这对加强鞘翅目害虫的检疫与防治有着积极意义。更多还原

【Abstract】 Beetles are insects of the order Coleoptera, the largest order in the animal kingdom, with over 250,000 species. They are one of the most common insects and exist in every variety of habitat except the oceans and polar regions. Some beetles feed on the leaves of a number of garden crops, flowers,trees and shrubs. Some beetles trasmit viral, bacterial, or fungal agents that cause plant desease. And some beetles may damage a wide variety of stored materials: stored wood, fabrics made of animal materials and stored food. Beetle insects present greate threat to agriculture and forest production. Correct and timely identification for these insects has important significance. For many years it has been performed by human experts with rich experience. However, sufficient number of competent human experts are not available to cover the large area. This paper present the design and implementation of an indentification systerm for the automatic identificaton of the beetle pest images.The automatic identification of beetle pests using image processing system consists of three stages: (i)pest image processing, (ii)shape features extracting, and(iii) pest recognition. Two image acquisition systems were designed to capture the images with respect to insects’ length. Median filter is one of the non-linear filter methods, which can get rid of images’ noise effectevely and was used to enhance the pest images. Adaptive image segmentation based on threshold has been adopted to get the pest object. According to the insect taxonomic character, seven shape features have been extracted based on the segmentation result of the pest images, which are Area, Perimeter, X-Length,Y-Length, Eccentricity, Shape-Parameter and Circularity. These features are useful for beetle pests classification. Fuzzy classifier based on the fuzzy minimums and maximums rule, multi - class support vector classifier based on least squares support vector machines, neural networks classifier based on BP , were inroduced and built to indentify 34 species of beetle pests. Results show that classifier based on the fuzzy minimums and maximums rule can get a satisfactory recognition result. Using this classifier, the average recognition accuracy can reached 92.16%, and it is suitable for beetle pests indentification.The automatic identification system of beetle pest images based on VB and Matlab makes identification easier to people with less experience, whilst minimizing system complexity and implementing a user-friendly environment. The system could be recommended as a potentially valuable tool for beetle pest indentification.更多还原