【作者】 黄宏华；

【导师】 赵杰文； 蔡健荣；

【作者基本信息】 江苏大学 ， 农产品加工及贮藏工程， 2003， 硕士

【摘要】 家蚕微粒子病是一种古老的、分布很广、毁灭性强的蚕病。曾对蚕丝业造成很大的损失。而且该病是传染性蚕病中唯一受检疫的病种，属于国际蚕种检疫对象，被我国列入进境动物二类传染病。但自从巴斯德发明母蛾镜选法检查微粒子以来，100多年来，蚕种微粒子病检验一直沿用这种镜选技术。目前我国口岸系统主要采用镜选法检测。 母蛾镜选的检测方法，劳动强度很大、工作效率低，而且受人为影响大，客观性不强，也不符合当前社会法制化、标准化的要求。在这种情况下，本研究根据蚕业生产的特点和要求，运用图像处理技来实现机器的自动检测。 针对微粒子显微图像噪声大、对比度不高而且是小目标检测的情况，本研究首先增强图像的亮度和对比度；然后采用小波阈值去噪技术对图像进行滤波处理；结合微粒子形状特点采用灰度形态学方法实现了对微粒子与背景的分割，并运用二值形态滤波筛选出微粒子；根据微粒子的形态特点，提取出了周长、宽度、高度、面积、伸长度、形状复杂度和矩形度等7个形态特征参数。为了区分与微粒子形态类似的其他孢子，对微粒子图像提取了一个颜色特征；采用基于遗传算法的神经网络分类器对于微粒子进行了识别。 本研究对150个样本进行了试验，识别率达到89％，由于一个样本下需要拍摄多幅图像，所以对一个样本可以作出多次识别，这样可以提高识别的准确率，漏判率降低，整体识别率提高。 研究结果表明：计算机图像处理方法能够解决微粒子病的复现问题，以及提高了检验工作效率、降低了成本、减轻了劳动强度，对于家蚕微粒子检测具有很大的实用价值。更多还原

【Abstract】 Pebrine is a kind of ancient, wildly distributed and strong destructive silkworm disease. It has done great damage to the silk industry in history. Now, it has been the main disease of silkworm in our country. And Pebrine is the disease of silkworm uniquely quarantined in international market. Since Louis Pasteur developed the method to dectect the pebrine by microscopic a hundred years ago, it has been used exclusively, so do in the custom today. In this paper image recognition technique is attempted to detect pebrine disease to replace the behindhand microscopic method.Dectection by microscopic has many defects, for example, it results in physically or mentally fatigued of human inspectors, so it can not guarantee the objectivity. And the dectection can not always implemented with great efficiency and high speed ,and so it can not meet the requirement of social legal system and standardization.In this research, according to the character of the micrograph, the original image is enhanced in brightness and contrast .Based on the threshold ,a new filtering algorithm with multiwavelet was proposed. And the mathematical morphology is considered as a effective way being applied for segmenting the pebrine and background according to the local gray feature. A mass of noises and small impurities are filtered by morphology filtering method and the preparatory separation of pebrine with impurities is realized. Eight parameters, such as perimeter, area, width, height, roundness, elongation-rate, complexity and color feature are extracted to remove the remainder impurities . Then pebrine was recognized and classified by neural network based on genetic algorithm.150 pebrine images have been tested in the experiment. The complete recognition exactness is 89%.This work is benefit to detecting the pebrine and achieve a good result.更多还原