【作者】 芦亚亚；

【导师】 古辉；

【作者基本信息】 浙江工业大学 ， 计算机应用技术， 2007， 硕士

【摘要】 果蔬采收是一项劳动密集型的工作，在很多国家，随着劳动力的高龄化和人力资源的缺乏，人工采收的成本在果蔬的整个生产成本中占了很大的比例。而我国是一个农业大国，果蔬产量多，品种丰富，出口量大。但是，现阶段的采收基本依靠人力，成本高、效率低。因此，实现果蔬采收的自动化迫在眉睫。利用计算机视觉系统对自然场景下的果蔬进行分割识别是实现机器人自动化采摘关键的一步。但是，自然场景下生长的果蔬具有极高的随机性，要实现一个完整的果蔬识别系统，必须解决六大问题，包括实现果实在自然光照条件下、阴影下、被遮挡情况下以及与背景色相似情况下的有效识别；解决果实中心点和采摘点的准确定位问题。针对上述六大难题，本文主要的工作有：(1)根据各颜色模型的特点，提出了两种新的颜色模型LHM和YNM。在分析其实验结果后选用LHM模型作为本文的颜色模型。依据分类识别的原理和LHM的聚类性成功解决了被遮挡、自然光照下和阴影下的果蔬识别问题。(2)研究和利用灰度共生矩阵提取纹理特征，得到可区分果实和背景的两大纹理特征：熵和能量。然后，综合颜色特征和纹理特征，解决了果实和背景颜色相似的识别问题。(3)构造了果蔬模型，解决了定位果蔬中心的难点；给出了几何校正和果蔬采摘的新理念，解决了被遮挡的、自然下垂生长的和非自然下垂生长的果蔬采摘点的定位问题。更多还原

【Abstract】 Fruit harvesting is a labor-concentrated job. In many countries, withthe senility and lack of labor, the cost of manual harvesting takes a lion’sshare within the total cost of yielding the fruit and vegetables. Yet, ourcountry is a great agricultural country, is rich in yield, variety and exportquantity. But generally speaking, nowadays, the harvesting of the fruitand vegetables mostly depends on the manpower that is high cost but lowefficiency. So, the realization of automatic harvesting is extremely urgent.It is a crucial step in realizing the robotic harvesting by using thecomputer vision to segment and recognize the fruit under nature scenes.However, due to the fact that the growth of the fruit under nature scenesbear high randomcity, thus, in order to implement an intact recognitionsystem, we need to resolve six problems including the efficientrecognition of the fruit object under nature illumination, in shadows, when the fruit object partially occluded or similar background colorpossessed; as well as the precise location of the fruit object center andabscission point.Aimed at the puzzles mentioned above, the main tasks we have doneare as follows:(1) Put forward two new color models LHM and YNM according tothe characteristics of each model, adopted the LHM model afteranalyzing the experimental result. By the principal of the classificationand the clustering characteristic of LHM, we settled the recognition of thefruit partially occluded, under nature illumination or shadow successfully.(2) Got two texture features: entropy and energy which candistinguish the fruit and the background by making use of the Gray levelco-occurrence matrix. Integrating the color and texture features as thefollowing step, we solved the recognition of the fruit whose color issimilar to the background.(3) Constructed geometry models for the fruit and vegetables, workout the difficulty of locating the center of the fruit; Brought forward thenovel conception of geometrical emendation and fruit harvesting,resolved the location of the fruit which covered the situation that the fruitare partially occluded, develop naturally hang and unnaturally hang.更多还原