【作者】 王辉；

【导师】 张小超； Qin.Zhang；

【作者基本信息】 中国农业机械化科学研究院 ， 机械设计及理论， 2011， 博士

【摘要】 针对目前果园种植和管理技术落后、果园劳动强度大、水果品质落后、农药喷洒过多等问题,采用机器视觉技术,研究了水果外观品质分级方法、根蘖对靶喷药技术和水果采摘机器人技术,所研究的方法提升了果园的自动化程度,既能减轻劳动强度,又能抢农时,减少损失,促进果品优质高产。在总结国内外相关研究的基础上,本文主要研究了收获前检测樱桃外观品质的方法,葡萄根蘖的识别和定位方法以及树上苹果的空间识别与定位方法：1、研究了自然环境下樱桃表面颜色分级的模式识别方法。首次建立了樱桃表面均一颜色在同种光照环境下变化的数学模型,并依据此模型建立了樱桃表面颜色模式识别分类器,解决了樱桃表面的颜色受环境色温变化影响的问题。在色温分别为5700K(果园)、5400K和6500K(室内)三种光照条件下,樱桃分级处理试验的结果表明：果园环境下樱桃分级准确率达到87%,室内环境下樱桃分级准确率86.7%。2、研究了利用主动光源对樱桃表面颜色分级方法。应用主动光源有效去除樱桃表面自然光和镜面反射,提出了一种七等级樱桃表面颜色的图像处理算法。在三种光照条件下(阳光直射4800K,明亮阴影5700K,阴暗阴影7700K),试验结果明：樱桃颜色分级的准确率达到93%。3、研究了樱桃外径的测量方法。比较了椭圆拟合、圆拟合和旋转搜索三种方法测量樱桃外径的精确度,试验结果表明：椭圆拟合方法最有效,其标准偏差为0.64mm。4、研究了利用机器视觉系统和激光扫描雷达组合实现葡萄根蘖的实时检测和定位方法,开发了葡萄根蘖对靶喷药系统。研究了摄像机数学模型和自校正方法,首次提出了相机和激光扫描雷达组合测量空间位置和标定方法,研究了利用根蘖颜色、纹理和位置信息实现对根蘖的实时检测和定位的方法。在三种速度下(1.6、2.4、3.2km/h)的试验结果表明：根蘖实时识别与定位系统能快速精准的定位葡萄根蘖,葡萄茎秆的识别率达到98%以上,葡萄根蘖尺寸检测的平均精度为80%左右。5、建立了苹果采摘机器人的智能视觉系统,实现了机器人眼手空间信息的精准转换。首次提出了单双目组合式智能立体视觉系统,系统由红外与彩色双目立体视觉系统和近距离单目视觉系统组成。研究了基于红外与彩色立体视觉系统三维重建技术,在空间中利用颜色、深度和位置信息进行苹果识别的方法,多视觉传感器之间组合标定的方法。试验结果表明：在距离目标2.4m时,苹果识别率为83.3%,空间定位的标准偏差为4.9cm；在距离1.5m时,苹果识别率为100%,空间定位的标准偏差为3.3cm。更多还原

【Abstract】 Now, there are a lot of problems about orchard planting and management technology, such as labor-intensity, excessive pesticide spraying and low fruit quality and so on. Using machine vision technology to study fruit quality rating, target spraying and fruit picking robot, is significant to improve orchard automation technology, and reduce labor intensity, grab farming, reduce losses, promote high fruit quality. In summary, based on international research, this paper studies pre-harvest cherry quality detection methods, suckers real-time identification and location, and apple detection and location methods in space:1、Research on color rating for cherry fruits under natural lighting conditions. The changing mathematical model of cherry skin color is established firstly in the same light conditions. Based on the model a color classifier is established to solve the problem that cherry skin color affected by environmental color temperature. Validation test have been conducted under different lighting conditions (5700K in orchard, and 5400K,6500K in lab). The results showed that the method could automatically grade cherry samples into seven pre-defined color levels under varying lighting conditions. Using rating results obtained by an experienced horticulturist on same samples as bench marks, this developed computer rating method resulted in an acceptable accuracy level of 87% match.2、Using an active light source to conduct accurate color rating of sweet cherries in an outdoor environment. It used a camera flash to reduce the effects of two major obstacles in outdoor color rating:(1) inconsistent ambient light and (2) specular reflections on cherry skins. An image processing algorithm was developed to classify cheery colors into seven levels. Tests showed that the overall rating accuracy was 93% in three different outdoor lighting conditions(4800K,5700K, 7700K). The results validated the feasibility of using computer vision to realize accurate color rating of sweet cherries in an outdoor environment.3、Research on the measurement methods of cherry fruit width, including oval fitting, circle fitting and rotation searching three methods. Using cherry width results obtained by a ruler on same samples as bench marks, the results showed that the oval fitting method is better than others, the standard deviation (std) is 0.48mm, the coefficient of determination (R2) is 0.822.4、Developed the sucker location and recognition system based on a laser range finder and a color CCD camera. The camera mathematical model and self-calibration methods are studied. The space measurement and calibration method are developed firstly based on laser and camera. A real-time image processing algorithm was developed to detect and locate sucker based on sucker’s color, texture and position. The results of field trials showed that this method could locate and recognize suckers effectively. The measurement accuracy of sucker size is 80%, the identification accuracy of trunk is 98%.5、In order to auto pick apple, an intelligent vision system was design and implemented. A single and binocular stereo vision system was developed firstly, the system consists of a binocular vision system with infrared and color vision sensor and a single vision system to measure close object. A calibration with two vision system was done to ensure that the information can transmit each other accurately. Using three-dimensional reconstruction techniques, an image processing algorithm was developed to detect the apple, based on the color, depth shape and location. Validation test show that the accuracy of identification is 83.3%, the depth’s Mean Square Error (MSE) is 0.049m; when the distance is about 2.4 m from sensor to object, the accuracy is 100%, MSE is 2.3cm when the distance is about 1.5m. The results validated the feasibility of an intelligent vision system to realize accurate detection and location.更多还原